1fa9e4066Sahrens /* 2fa9e4066Sahrens * CDDL HEADER START 3fa9e4066Sahrens * 4fa9e4066Sahrens * The contents of this file are subject to the terms of the 5441d80aaSlling * Common Development and Distribution License (the "License"). 6441d80aaSlling * You may not use this file except in compliance with the License. 7fa9e4066Sahrens * 8fa9e4066Sahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9fa9e4066Sahrens * or http://www.opensolaris.org/os/licensing. 10fa9e4066Sahrens * See the License for the specific language governing permissions 11fa9e4066Sahrens * and limitations under the License. 12fa9e4066Sahrens * 13fa9e4066Sahrens * When distributing Covered Code, include this CDDL HEADER in each 14fa9e4066Sahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15fa9e4066Sahrens * If applicable, add the following below this CDDL HEADER, with the 16fa9e4066Sahrens * fields enclosed by brackets "[]" replaced with your own identifying 17fa9e4066Sahrens * information: Portions Copyright [yyyy] [name of copyright owner] 18fa9e4066Sahrens * 19fa9e4066Sahrens * CDDL HEADER END 20fa9e4066Sahrens */ 2199653d4eSeschrock 22fa9e4066Sahrens /* 2398d1cbfeSGeorge Wilson * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 245cabbc6bSPrashanth Sreenivasa * Copyright (c) 2011, 2018 by Delphix. All rights reserved. 255f368aefSYuri Pankov * Copyright 2017 Nexenta Systems, Inc. 26c3d26abcSMatthew Ahrens * Copyright (c) 2014 Integros [integros.com] 27c8811bd3SToomas Soome * Copyright 2016 Toomas Soome <tsoome@me.com> 28ce1577b0SDave Eddy * Copyright 2017 Joyent, Inc. 29fa9e4066Sahrens */ 30fa9e4066Sahrens 31fa9e4066Sahrens #include <sys/zfs_context.h> 32ea8dc4b6Seschrock #include <sys/fm/fs/zfs.h> 33fa9e4066Sahrens #include <sys/spa.h> 34fa9e4066Sahrens #include <sys/spa_impl.h> 355cabbc6bSPrashanth Sreenivasa #include <sys/bpobj.h> 36fa9e4066Sahrens #include <sys/dmu.h> 37fa9e4066Sahrens #include <sys/dmu_tx.h> 385cabbc6bSPrashanth Sreenivasa #include <sys/dsl_dir.h> 39fa9e4066Sahrens #include <sys/vdev_impl.h> 40fa9e4066Sahrens #include <sys/uberblock_impl.h> 41fa9e4066Sahrens #include <sys/metaslab.h> 42fa9e4066Sahrens #include <sys/metaslab_impl.h> 43fa9e4066Sahrens #include <sys/space_map.h> 440713e232SGeorge Wilson #include <sys/space_reftree.h> 45fa9e4066Sahrens #include <sys/zio.h> 46fa9e4066Sahrens #include <sys/zap.h> 47fa9e4066Sahrens #include <sys/fs/zfs.h> 48c5904d13Seschrock #include <sys/arc.h> 49e6ca193dSGeorge Wilson #include <sys/zil.h> 503f9d6ad7SLin Ling #include <sys/dsl_scan.h> 51770499e1SDan Kimmel #include <sys/abd.h> 52fa9e4066Sahrens 53fa9e4066Sahrens /* 54fa9e4066Sahrens * Virtual device management. 55fa9e4066Sahrens */ 56fa9e4066Sahrens 57fa9e4066Sahrens static vdev_ops_t *vdev_ops_table[] = { 58fa9e4066Sahrens &vdev_root_ops, 59fa9e4066Sahrens &vdev_raidz_ops, 60fa9e4066Sahrens &vdev_mirror_ops, 61fa9e4066Sahrens &vdev_replacing_ops, 6299653d4eSeschrock &vdev_spare_ops, 63fa9e4066Sahrens &vdev_disk_ops, 64fa9e4066Sahrens &vdev_file_ops, 65fa9e4066Sahrens &vdev_missing_ops, 6688ecc943SGeorge Wilson &vdev_hole_ops, 675cabbc6bSPrashanth Sreenivasa &vdev_indirect_ops, 68fa9e4066Sahrens NULL 69fa9e4066Sahrens }; 70fa9e4066Sahrens 71088f3894Sahrens /* maximum scrub/resilver I/O queue per leaf vdev */ 72088f3894Sahrens int zfs_scrub_limit = 10; 7305b2b3b8Smishra 74*86714001SSerapheim Dimitropoulos /* maximum number of metaslabs per top-level vdev */ 75*86714001SSerapheim Dimitropoulos int vdev_max_ms_count = 200; 76*86714001SSerapheim Dimitropoulos 77*86714001SSerapheim Dimitropoulos /* minimum amount of metaslabs per top-level vdev */ 78*86714001SSerapheim Dimitropoulos int vdev_min_ms_count = 16; 79*86714001SSerapheim Dimitropoulos 80*86714001SSerapheim Dimitropoulos /* see comment in vdev_metaslab_set_size() */ 81*86714001SSerapheim Dimitropoulos int vdev_default_ms_shift = 29; 82*86714001SSerapheim Dimitropoulos 83*86714001SSerapheim Dimitropoulos boolean_t vdev_validate_skip = B_FALSE; 84*86714001SSerapheim Dimitropoulos 85bf3e216cSMatthew Ahrens /* 86*86714001SSerapheim Dimitropoulos * Since the DTL space map of a vdev is not expected to have a lot of 87*86714001SSerapheim Dimitropoulos * entries, we default its block size to 4K. 88bf3e216cSMatthew Ahrens */ 89*86714001SSerapheim Dimitropoulos int vdev_dtl_sm_blksz = (1 << 12); 90bf3e216cSMatthew Ahrens 91*86714001SSerapheim Dimitropoulos /* 92*86714001SSerapheim Dimitropoulos * vdev-wide space maps that have lots of entries written to them at 93*86714001SSerapheim Dimitropoulos * the end of each transaction can benefit from a higher I/O bandwidth 94*86714001SSerapheim Dimitropoulos * (e.g. vdev_obsolete_sm), thus we default their block size to 128K. 95*86714001SSerapheim Dimitropoulos */ 96*86714001SSerapheim Dimitropoulos int vdev_standard_sm_blksz = (1 << 17); 976f793812SPavel Zakharov 983ee8c80cSPavel Zakharov /*PRINTFLIKE2*/ 993ee8c80cSPavel Zakharov void 1003ee8c80cSPavel Zakharov vdev_dbgmsg(vdev_t *vd, const char *fmt, ...) 1013ee8c80cSPavel Zakharov { 1023ee8c80cSPavel Zakharov va_list adx; 1033ee8c80cSPavel Zakharov char buf[256]; 1043ee8c80cSPavel Zakharov 1053ee8c80cSPavel Zakharov va_start(adx, fmt); 1063ee8c80cSPavel Zakharov (void) vsnprintf(buf, sizeof (buf), fmt, adx); 1073ee8c80cSPavel Zakharov va_end(adx); 1083ee8c80cSPavel Zakharov 1093ee8c80cSPavel Zakharov if (vd->vdev_path != NULL) { 1103ee8c80cSPavel Zakharov zfs_dbgmsg("%s vdev '%s': %s", vd->vdev_ops->vdev_op_type, 1113ee8c80cSPavel Zakharov vd->vdev_path, buf); 1123ee8c80cSPavel Zakharov } else { 1133ee8c80cSPavel Zakharov zfs_dbgmsg("%s-%llu vdev (guid %llu): %s", 1143ee8c80cSPavel Zakharov vd->vdev_ops->vdev_op_type, 1153ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_id, 1163ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_guid, buf); 1173ee8c80cSPavel Zakharov } 1183ee8c80cSPavel Zakharov } 1193ee8c80cSPavel Zakharov 1206f793812SPavel Zakharov void 1216f793812SPavel Zakharov vdev_dbgmsg_print_tree(vdev_t *vd, int indent) 1226f793812SPavel Zakharov { 1236f793812SPavel Zakharov char state[20]; 1246f793812SPavel Zakharov 1256f793812SPavel Zakharov if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) { 1266f793812SPavel Zakharov zfs_dbgmsg("%*svdev %u: %s", indent, "", vd->vdev_id, 1276f793812SPavel Zakharov vd->vdev_ops->vdev_op_type); 1286f793812SPavel Zakharov return; 1296f793812SPavel Zakharov } 1306f793812SPavel Zakharov 1316f793812SPavel Zakharov switch (vd->vdev_state) { 1326f793812SPavel Zakharov case VDEV_STATE_UNKNOWN: 1336f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "unknown"); 1346f793812SPavel Zakharov break; 1356f793812SPavel Zakharov case VDEV_STATE_CLOSED: 1366f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "closed"); 1376f793812SPavel Zakharov break; 1386f793812SPavel Zakharov case VDEV_STATE_OFFLINE: 1396f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "offline"); 1406f793812SPavel Zakharov break; 1416f793812SPavel Zakharov case VDEV_STATE_REMOVED: 1426f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "removed"); 1436f793812SPavel Zakharov break; 1446f793812SPavel Zakharov case VDEV_STATE_CANT_OPEN: 1456f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "can't open"); 1466f793812SPavel Zakharov break; 1476f793812SPavel Zakharov case VDEV_STATE_FAULTED: 1486f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "faulted"); 1496f793812SPavel Zakharov break; 1506f793812SPavel Zakharov case VDEV_STATE_DEGRADED: 1516f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "degraded"); 1526f793812SPavel Zakharov break; 1536f793812SPavel Zakharov case VDEV_STATE_HEALTHY: 1546f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "healthy"); 1556f793812SPavel Zakharov break; 1566f793812SPavel Zakharov default: 1576f793812SPavel Zakharov (void) snprintf(state, sizeof (state), "<state %u>", 1586f793812SPavel Zakharov (uint_t)vd->vdev_state); 1596f793812SPavel Zakharov } 1606f793812SPavel Zakharov 1616f793812SPavel Zakharov zfs_dbgmsg("%*svdev %u: %s%s, guid: %llu, path: %s, %s", indent, 1626f793812SPavel Zakharov "", vd->vdev_id, vd->vdev_ops->vdev_op_type, 1636f793812SPavel Zakharov vd->vdev_islog ? " (log)" : "", 1646f793812SPavel Zakharov (u_longlong_t)vd->vdev_guid, 1656f793812SPavel Zakharov vd->vdev_path ? vd->vdev_path : "N/A", state); 1666f793812SPavel Zakharov 1676f793812SPavel Zakharov for (uint64_t i = 0; i < vd->vdev_children; i++) 1686f793812SPavel Zakharov vdev_dbgmsg_print_tree(vd->vdev_child[i], indent + 2); 1696f793812SPavel Zakharov } 1706f793812SPavel Zakharov 171fa9e4066Sahrens /* 172fa9e4066Sahrens * Given a vdev type, return the appropriate ops vector. 173fa9e4066Sahrens */ 174fa9e4066Sahrens static vdev_ops_t * 175fa9e4066Sahrens vdev_getops(const char *type) 176fa9e4066Sahrens { 177fa9e4066Sahrens vdev_ops_t *ops, **opspp; 178fa9e4066Sahrens 179fa9e4066Sahrens for (opspp = vdev_ops_table; (ops = *opspp) != NULL; opspp++) 180fa9e4066Sahrens if (strcmp(ops->vdev_op_type, type) == 0) 181fa9e4066Sahrens break; 182fa9e4066Sahrens 183fa9e4066Sahrens return (ops); 184fa9e4066Sahrens } 185fa9e4066Sahrens 186fa9e4066Sahrens /* 187fa9e4066Sahrens * Default asize function: return the MAX of psize with the asize of 188fa9e4066Sahrens * all children. This is what's used by anything other than RAID-Z. 189fa9e4066Sahrens */ 190fa9e4066Sahrens uint64_t 191fa9e4066Sahrens vdev_default_asize(vdev_t *vd, uint64_t psize) 192fa9e4066Sahrens { 193ecc2d604Sbonwick uint64_t asize = P2ROUNDUP(psize, 1ULL << vd->vdev_top->vdev_ashift); 194fa9e4066Sahrens uint64_t csize; 195fa9e4066Sahrens 196573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 197fa9e4066Sahrens csize = vdev_psize_to_asize(vd->vdev_child[c], psize); 198fa9e4066Sahrens asize = MAX(asize, csize); 199fa9e4066Sahrens } 200fa9e4066Sahrens 201fa9e4066Sahrens return (asize); 202fa9e4066Sahrens } 203fa9e4066Sahrens 2042a79c5feSlling /* 205573ca77eSGeorge Wilson * Get the minimum allocatable size. We define the allocatable size as 206573ca77eSGeorge Wilson * the vdev's asize rounded to the nearest metaslab. This allows us to 207573ca77eSGeorge Wilson * replace or attach devices which don't have the same physical size but 208573ca77eSGeorge Wilson * can still satisfy the same number of allocations. 2092a79c5feSlling */ 2102a79c5feSlling uint64_t 211573ca77eSGeorge Wilson vdev_get_min_asize(vdev_t *vd) 2122a79c5feSlling { 213573ca77eSGeorge Wilson vdev_t *pvd = vd->vdev_parent; 2142a79c5feSlling 215573ca77eSGeorge Wilson /* 2164263d13fSGeorge Wilson * If our parent is NULL (inactive spare or cache) or is the root, 217573ca77eSGeorge Wilson * just return our own asize. 218573ca77eSGeorge Wilson */ 219573ca77eSGeorge Wilson if (pvd == NULL) 220573ca77eSGeorge Wilson return (vd->vdev_asize); 2212a79c5feSlling 2222a79c5feSlling /* 223573ca77eSGeorge Wilson * The top-level vdev just returns the allocatable size rounded 224573ca77eSGeorge Wilson * to the nearest metaslab. 2252a79c5feSlling */ 226573ca77eSGeorge Wilson if (vd == vd->vdev_top) 227573ca77eSGeorge Wilson return (P2ALIGN(vd->vdev_asize, 1ULL << vd->vdev_ms_shift)); 2282a79c5feSlling 229573ca77eSGeorge Wilson /* 230573ca77eSGeorge Wilson * The allocatable space for a raidz vdev is N * sizeof(smallest child), 231573ca77eSGeorge Wilson * so each child must provide at least 1/Nth of its asize. 232573ca77eSGeorge Wilson */ 233573ca77eSGeorge Wilson if (pvd->vdev_ops == &vdev_raidz_ops) 234c040c10cSSteven Hartland return ((pvd->vdev_min_asize + pvd->vdev_children - 1) / 235c040c10cSSteven Hartland pvd->vdev_children); 2362a79c5feSlling 237573ca77eSGeorge Wilson return (pvd->vdev_min_asize); 238573ca77eSGeorge Wilson } 2392a79c5feSlling 240573ca77eSGeorge Wilson void 241573ca77eSGeorge Wilson vdev_set_min_asize(vdev_t *vd) 242573ca77eSGeorge Wilson { 243573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 244573ca77eSGeorge Wilson 245573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 246573ca77eSGeorge Wilson vdev_set_min_asize(vd->vdev_child[c]); 2472a79c5feSlling } 2482a79c5feSlling 249fa9e4066Sahrens vdev_t * 250fa9e4066Sahrens vdev_lookup_top(spa_t *spa, uint64_t vdev) 251fa9e4066Sahrens { 252fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 253fa9e4066Sahrens 254e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 255e05725b1Sbonwick 256088f3894Sahrens if (vdev < rvd->vdev_children) { 257088f3894Sahrens ASSERT(rvd->vdev_child[vdev] != NULL); 258fa9e4066Sahrens return (rvd->vdev_child[vdev]); 259088f3894Sahrens } 260fa9e4066Sahrens 261fa9e4066Sahrens return (NULL); 262fa9e4066Sahrens } 263fa9e4066Sahrens 264fa9e4066Sahrens vdev_t * 265fa9e4066Sahrens vdev_lookup_by_guid(vdev_t *vd, uint64_t guid) 266fa9e4066Sahrens { 267fa9e4066Sahrens vdev_t *mvd; 268fa9e4066Sahrens 2690e34b6a7Sbonwick if (vd->vdev_guid == guid) 270fa9e4066Sahrens return (vd); 271fa9e4066Sahrens 272573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 273fa9e4066Sahrens if ((mvd = vdev_lookup_by_guid(vd->vdev_child[c], guid)) != 274fa9e4066Sahrens NULL) 275fa9e4066Sahrens return (mvd); 276fa9e4066Sahrens 277fa9e4066Sahrens return (NULL); 278fa9e4066Sahrens } 279fa9e4066Sahrens 28012380e1eSArne Jansen static int 28112380e1eSArne Jansen vdev_count_leaves_impl(vdev_t *vd) 28212380e1eSArne Jansen { 28312380e1eSArne Jansen int n = 0; 28412380e1eSArne Jansen 28512380e1eSArne Jansen if (vd->vdev_ops->vdev_op_leaf) 28612380e1eSArne Jansen return (1); 28712380e1eSArne Jansen 28812380e1eSArne Jansen for (int c = 0; c < vd->vdev_children; c++) 28912380e1eSArne Jansen n += vdev_count_leaves_impl(vd->vdev_child[c]); 29012380e1eSArne Jansen 29112380e1eSArne Jansen return (n); 29212380e1eSArne Jansen } 29312380e1eSArne Jansen 29412380e1eSArne Jansen int 29512380e1eSArne Jansen vdev_count_leaves(spa_t *spa) 29612380e1eSArne Jansen { 29712380e1eSArne Jansen return (vdev_count_leaves_impl(spa->spa_root_vdev)); 29812380e1eSArne Jansen } 29912380e1eSArne Jansen 300fa9e4066Sahrens void 301fa9e4066Sahrens vdev_add_child(vdev_t *pvd, vdev_t *cvd) 302fa9e4066Sahrens { 303fa9e4066Sahrens size_t oldsize, newsize; 304fa9e4066Sahrens uint64_t id = cvd->vdev_id; 305fa9e4066Sahrens vdev_t **newchild; 30681cd5c55SMatthew Ahrens spa_t *spa = cvd->vdev_spa; 307fa9e4066Sahrens 30881cd5c55SMatthew Ahrens ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 309fa9e4066Sahrens ASSERT(cvd->vdev_parent == NULL); 310fa9e4066Sahrens 311fa9e4066Sahrens cvd->vdev_parent = pvd; 312fa9e4066Sahrens 313fa9e4066Sahrens if (pvd == NULL) 314fa9e4066Sahrens return; 315fa9e4066Sahrens 316fa9e4066Sahrens ASSERT(id >= pvd->vdev_children || pvd->vdev_child[id] == NULL); 317fa9e4066Sahrens 318fa9e4066Sahrens oldsize = pvd->vdev_children * sizeof (vdev_t *); 319fa9e4066Sahrens pvd->vdev_children = MAX(pvd->vdev_children, id + 1); 320fa9e4066Sahrens newsize = pvd->vdev_children * sizeof (vdev_t *); 321fa9e4066Sahrens 322fa9e4066Sahrens newchild = kmem_zalloc(newsize, KM_SLEEP); 323fa9e4066Sahrens if (pvd->vdev_child != NULL) { 324fa9e4066Sahrens bcopy(pvd->vdev_child, newchild, oldsize); 325fa9e4066Sahrens kmem_free(pvd->vdev_child, oldsize); 326fa9e4066Sahrens } 327fa9e4066Sahrens 328fa9e4066Sahrens pvd->vdev_child = newchild; 329fa9e4066Sahrens pvd->vdev_child[id] = cvd; 330fa9e4066Sahrens 331fa9e4066Sahrens cvd->vdev_top = (pvd->vdev_top ? pvd->vdev_top: cvd); 332fa9e4066Sahrens ASSERT(cvd->vdev_top->vdev_parent->vdev_parent == NULL); 333fa9e4066Sahrens 334fa9e4066Sahrens /* 335fa9e4066Sahrens * Walk up all ancestors to update guid sum. 336fa9e4066Sahrens */ 337fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 338fa9e4066Sahrens pvd->vdev_guid_sum += cvd->vdev_guid_sum; 339fa9e4066Sahrens } 340fa9e4066Sahrens 341fa9e4066Sahrens void 342fa9e4066Sahrens vdev_remove_child(vdev_t *pvd, vdev_t *cvd) 343fa9e4066Sahrens { 344fa9e4066Sahrens int c; 345fa9e4066Sahrens uint_t id = cvd->vdev_id; 346fa9e4066Sahrens 347fa9e4066Sahrens ASSERT(cvd->vdev_parent == pvd); 348fa9e4066Sahrens 349fa9e4066Sahrens if (pvd == NULL) 350fa9e4066Sahrens return; 351fa9e4066Sahrens 352fa9e4066Sahrens ASSERT(id < pvd->vdev_children); 353fa9e4066Sahrens ASSERT(pvd->vdev_child[id] == cvd); 354fa9e4066Sahrens 355fa9e4066Sahrens pvd->vdev_child[id] = NULL; 356fa9e4066Sahrens cvd->vdev_parent = NULL; 357fa9e4066Sahrens 358fa9e4066Sahrens for (c = 0; c < pvd->vdev_children; c++) 359fa9e4066Sahrens if (pvd->vdev_child[c]) 360fa9e4066Sahrens break; 361fa9e4066Sahrens 362fa9e4066Sahrens if (c == pvd->vdev_children) { 363fa9e4066Sahrens kmem_free(pvd->vdev_child, c * sizeof (vdev_t *)); 364fa9e4066Sahrens pvd->vdev_child = NULL; 365fa9e4066Sahrens pvd->vdev_children = 0; 366fa9e4066Sahrens } 367fa9e4066Sahrens 368fa9e4066Sahrens /* 369fa9e4066Sahrens * Walk up all ancestors to update guid sum. 370fa9e4066Sahrens */ 371fa9e4066Sahrens for (; pvd != NULL; pvd = pvd->vdev_parent) 372fa9e4066Sahrens pvd->vdev_guid_sum -= cvd->vdev_guid_sum; 373fa9e4066Sahrens } 374fa9e4066Sahrens 375fa9e4066Sahrens /* 376fa9e4066Sahrens * Remove any holes in the child array. 377fa9e4066Sahrens */ 378fa9e4066Sahrens void 379fa9e4066Sahrens vdev_compact_children(vdev_t *pvd) 380fa9e4066Sahrens { 381fa9e4066Sahrens vdev_t **newchild, *cvd; 382fa9e4066Sahrens int oldc = pvd->vdev_children; 383573ca77eSGeorge Wilson int newc; 384fa9e4066Sahrens 385e14bb325SJeff Bonwick ASSERT(spa_config_held(pvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 386fa9e4066Sahrens 387573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) 388fa9e4066Sahrens if (pvd->vdev_child[c]) 389fa9e4066Sahrens newc++; 390fa9e4066Sahrens 391fa9e4066Sahrens newchild = kmem_alloc(newc * sizeof (vdev_t *), KM_SLEEP); 392fa9e4066Sahrens 393573ca77eSGeorge Wilson for (int c = newc = 0; c < oldc; c++) { 394fa9e4066Sahrens if ((cvd = pvd->vdev_child[c]) != NULL) { 395fa9e4066Sahrens newchild[newc] = cvd; 396fa9e4066Sahrens cvd->vdev_id = newc++; 397fa9e4066Sahrens } 398fa9e4066Sahrens } 399fa9e4066Sahrens 400fa9e4066Sahrens kmem_free(pvd->vdev_child, oldc * sizeof (vdev_t *)); 401fa9e4066Sahrens pvd->vdev_child = newchild; 402fa9e4066Sahrens pvd->vdev_children = newc; 403fa9e4066Sahrens } 404fa9e4066Sahrens 405fa9e4066Sahrens /* 406fa9e4066Sahrens * Allocate and minimally initialize a vdev_t. 407fa9e4066Sahrens */ 40888ecc943SGeorge Wilson vdev_t * 409fa9e4066Sahrens vdev_alloc_common(spa_t *spa, uint_t id, uint64_t guid, vdev_ops_t *ops) 410fa9e4066Sahrens { 411fa9e4066Sahrens vdev_t *vd; 4125cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic; 413fa9e4066Sahrens 414fa9e4066Sahrens vd = kmem_zalloc(sizeof (vdev_t), KM_SLEEP); 4155cabbc6bSPrashanth Sreenivasa vic = &vd->vdev_indirect_config; 416fa9e4066Sahrens 4170e34b6a7Sbonwick if (spa->spa_root_vdev == NULL) { 4180e34b6a7Sbonwick ASSERT(ops == &vdev_root_ops); 4190e34b6a7Sbonwick spa->spa_root_vdev = vd; 420e9103aaeSGarrett D'Amore spa->spa_load_guid = spa_generate_guid(NULL); 4210e34b6a7Sbonwick } 4220e34b6a7Sbonwick 42388ecc943SGeorge Wilson if (guid == 0 && ops != &vdev_hole_ops) { 4240e34b6a7Sbonwick if (spa->spa_root_vdev == vd) { 4250e34b6a7Sbonwick /* 4260e34b6a7Sbonwick * The root vdev's guid will also be the pool guid, 4270e34b6a7Sbonwick * which must be unique among all pools. 4280e34b6a7Sbonwick */ 4291195e687SMark J Musante guid = spa_generate_guid(NULL); 4300e34b6a7Sbonwick } else { 4310e34b6a7Sbonwick /* 4320e34b6a7Sbonwick * Any other vdev's guid must be unique within the pool. 4330e34b6a7Sbonwick */ 4341195e687SMark J Musante guid = spa_generate_guid(spa); 4350e34b6a7Sbonwick } 4360e34b6a7Sbonwick ASSERT(!spa_guid_exists(spa_guid(spa), guid)); 4370e34b6a7Sbonwick } 4380e34b6a7Sbonwick 439fa9e4066Sahrens vd->vdev_spa = spa; 440fa9e4066Sahrens vd->vdev_id = id; 441fa9e4066Sahrens vd->vdev_guid = guid; 442fa9e4066Sahrens vd->vdev_guid_sum = guid; 443fa9e4066Sahrens vd->vdev_ops = ops; 444fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 44588ecc943SGeorge Wilson vd->vdev_ishole = (ops == &vdev_hole_ops); 4465cabbc6bSPrashanth Sreenivasa vic->vic_prev_indirect_vdev = UINT64_MAX; 4475cabbc6bSPrashanth Sreenivasa 4485cabbc6bSPrashanth Sreenivasa rw_init(&vd->vdev_indirect_rwlock, NULL, RW_DEFAULT, NULL); 4495cabbc6bSPrashanth Sreenivasa mutex_init(&vd->vdev_obsolete_lock, NULL, MUTEX_DEFAULT, NULL); 4505cabbc6bSPrashanth Sreenivasa vd->vdev_obsolete_segments = range_tree_create(NULL, NULL); 451fa9e4066Sahrens 452fa9e4066Sahrens mutex_init(&vd->vdev_dtl_lock, NULL, MUTEX_DEFAULT, NULL); 4535ad82045Snd mutex_init(&vd->vdev_stat_lock, NULL, MUTEX_DEFAULT, NULL); 454e14bb325SJeff Bonwick mutex_init(&vd->vdev_probe_lock, NULL, MUTEX_DEFAULT, NULL); 4550f7643c7SGeorge Wilson mutex_init(&vd->vdev_queue_lock, NULL, MUTEX_DEFAULT, NULL); 4568ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 4575cabbc6bSPrashanth Sreenivasa vd->vdev_dtl[t] = range_tree_create(NULL, NULL); 4588ad4d6ddSJeff Bonwick } 459b7b2590dSMatthew Ahrens txg_list_create(&vd->vdev_ms_list, spa, 460fa9e4066Sahrens offsetof(struct metaslab, ms_txg_node)); 461b7b2590dSMatthew Ahrens txg_list_create(&vd->vdev_dtl_list, spa, 462fa9e4066Sahrens offsetof(struct vdev, vdev_dtl_node)); 463fa9e4066Sahrens vd->vdev_stat.vs_timestamp = gethrtime(); 4643d7072f8Seschrock vdev_queue_init(vd); 4653d7072f8Seschrock vdev_cache_init(vd); 466fa9e4066Sahrens 467fa9e4066Sahrens return (vd); 468fa9e4066Sahrens } 469fa9e4066Sahrens 470fa9e4066Sahrens /* 471fa9e4066Sahrens * Allocate a new vdev. The 'alloctype' is used to control whether we are 472fa9e4066Sahrens * creating a new vdev or loading an existing one - the behavior is slightly 473fa9e4066Sahrens * different for each case. 474fa9e4066Sahrens */ 47599653d4eSeschrock int 47699653d4eSeschrock vdev_alloc(spa_t *spa, vdev_t **vdp, nvlist_t *nv, vdev_t *parent, uint_t id, 47799653d4eSeschrock int alloctype) 478fa9e4066Sahrens { 479fa9e4066Sahrens vdev_ops_t *ops; 480fa9e4066Sahrens char *type; 4818654d025Sperrin uint64_t guid = 0, islog, nparity; 482fa9e4066Sahrens vdev_t *vd; 4835cabbc6bSPrashanth Sreenivasa vdev_indirect_config_t *vic; 484fa9e4066Sahrens 485e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 486fa9e4066Sahrens 487fa9e4066Sahrens if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0) 488be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 489fa9e4066Sahrens 490fa9e4066Sahrens if ((ops = vdev_getops(type)) == NULL) 491be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 492fa9e4066Sahrens 493fa9e4066Sahrens /* 494fa9e4066Sahrens * If this is a load, get the vdev guid from the nvlist. 495fa9e4066Sahrens * Otherwise, vdev_alloc_common() will generate one for us. 496fa9e4066Sahrens */ 497fa9e4066Sahrens if (alloctype == VDEV_ALLOC_LOAD) { 498fa9e4066Sahrens uint64_t label_id; 499fa9e4066Sahrens 500fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ID, &label_id) || 501fa9e4066Sahrens label_id != id) 502be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 503fa9e4066Sahrens 504fa9e4066Sahrens if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 505be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 50699653d4eSeschrock } else if (alloctype == VDEV_ALLOC_SPARE) { 50799653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 508be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 509fa94a07fSbrendan } else if (alloctype == VDEV_ALLOC_L2CACHE) { 510fa94a07fSbrendan if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 511be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 51221ecdf64SLin Ling } else if (alloctype == VDEV_ALLOC_ROOTPOOL) { 51321ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_GUID, &guid) != 0) 514be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 515fa9e4066Sahrens } 516fa9e4066Sahrens 51799653d4eSeschrock /* 51899653d4eSeschrock * The first allocated vdev must be of type 'root'. 51999653d4eSeschrock */ 52099653d4eSeschrock if (ops != &vdev_root_ops && spa->spa_root_vdev == NULL) 521be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 52299653d4eSeschrock 5238654d025Sperrin /* 5248654d025Sperrin * Determine whether we're a log vdev. 5258654d025Sperrin */ 5268654d025Sperrin islog = 0; 5278654d025Sperrin (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_LOG, &islog); 528990b4856Slling if (islog && spa_version(spa) < SPA_VERSION_SLOGS) 529be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 530fa9e4066Sahrens 53188ecc943SGeorge Wilson if (ops == &vdev_hole_ops && spa_version(spa) < SPA_VERSION_HOLES) 532be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 53388ecc943SGeorge Wilson 53499653d4eSeschrock /* 5358654d025Sperrin * Set the nparity property for RAID-Z vdevs. 53699653d4eSeschrock */ 5378654d025Sperrin nparity = -1ULL; 53899653d4eSeschrock if (ops == &vdev_raidz_ops) { 53999653d4eSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NPARITY, 5408654d025Sperrin &nparity) == 0) { 541b24ab676SJeff Bonwick if (nparity == 0 || nparity > VDEV_RAIDZ_MAXPARITY) 542be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 54399653d4eSeschrock /* 544f94275ceSAdam Leventhal * Previous versions could only support 1 or 2 parity 545f94275ceSAdam Leventhal * device. 54699653d4eSeschrock */ 547f94275ceSAdam Leventhal if (nparity > 1 && 548f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ2) 549be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 550f94275ceSAdam Leventhal if (nparity > 2 && 551f94275ceSAdam Leventhal spa_version(spa) < SPA_VERSION_RAIDZ3) 552be6fd75aSMatthew Ahrens return (SET_ERROR(ENOTSUP)); 55399653d4eSeschrock } else { 55499653d4eSeschrock /* 55599653d4eSeschrock * We require the parity to be specified for SPAs that 55699653d4eSeschrock * support multiple parity levels. 55799653d4eSeschrock */ 558f94275ceSAdam Leventhal if (spa_version(spa) >= SPA_VERSION_RAIDZ2) 559be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 56099653d4eSeschrock /* 56199653d4eSeschrock * Otherwise, we default to 1 parity device for RAID-Z. 56299653d4eSeschrock */ 5638654d025Sperrin nparity = 1; 56499653d4eSeschrock } 56599653d4eSeschrock } else { 5668654d025Sperrin nparity = 0; 56799653d4eSeschrock } 5688654d025Sperrin ASSERT(nparity != -1ULL); 5698654d025Sperrin 5708654d025Sperrin vd = vdev_alloc_common(spa, id, guid, ops); 5715cabbc6bSPrashanth Sreenivasa vic = &vd->vdev_indirect_config; 5728654d025Sperrin 5738654d025Sperrin vd->vdev_islog = islog; 5748654d025Sperrin vd->vdev_nparity = nparity; 5758654d025Sperrin 5768654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &vd->vdev_path) == 0) 5778654d025Sperrin vd->vdev_path = spa_strdup(vd->vdev_path); 5788654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_DEVID, &vd->vdev_devid) == 0) 5798654d025Sperrin vd->vdev_devid = spa_strdup(vd->vdev_devid); 5808654d025Sperrin if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PHYS_PATH, 5818654d025Sperrin &vd->vdev_physpath) == 0) 5828654d025Sperrin vd->vdev_physpath = spa_strdup(vd->vdev_physpath); 5836809eb4eSEric Schrock if (nvlist_lookup_string(nv, ZPOOL_CONFIG_FRU, &vd->vdev_fru) == 0) 5846809eb4eSEric Schrock vd->vdev_fru = spa_strdup(vd->vdev_fru); 58599653d4eSeschrock 586afefbcddSeschrock /* 587afefbcddSeschrock * Set the whole_disk property. If it's not specified, leave the value 588afefbcddSeschrock * as -1. 589afefbcddSeschrock */ 590afefbcddSeschrock if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, 591afefbcddSeschrock &vd->vdev_wholedisk) != 0) 592afefbcddSeschrock vd->vdev_wholedisk = -1ULL; 593afefbcddSeschrock 5945cabbc6bSPrashanth Sreenivasa ASSERT0(vic->vic_mapping_object); 5955cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_OBJECT, 5965cabbc6bSPrashanth Sreenivasa &vic->vic_mapping_object); 5975cabbc6bSPrashanth Sreenivasa ASSERT0(vic->vic_births_object); 5985cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_INDIRECT_BIRTHS, 5995cabbc6bSPrashanth Sreenivasa &vic->vic_births_object); 6005cabbc6bSPrashanth Sreenivasa ASSERT3U(vic->vic_prev_indirect_vdev, ==, UINT64_MAX); 6015cabbc6bSPrashanth Sreenivasa (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_PREV_INDIRECT_VDEV, 6025cabbc6bSPrashanth Sreenivasa &vic->vic_prev_indirect_vdev); 6035cabbc6bSPrashanth Sreenivasa 604ea8dc4b6Seschrock /* 605ea8dc4b6Seschrock * Look for the 'not present' flag. This will only be set if the device 606ea8dc4b6Seschrock * was not present at the time of import. 607ea8dc4b6Seschrock */ 6086809eb4eSEric Schrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_NOT_PRESENT, 6096809eb4eSEric Schrock &vd->vdev_not_present); 610ea8dc4b6Seschrock 611ecc2d604Sbonwick /* 612ecc2d604Sbonwick * Get the alignment requirement. 613ecc2d604Sbonwick */ 614ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASHIFT, &vd->vdev_ashift); 615ecc2d604Sbonwick 61688ecc943SGeorge Wilson /* 61788ecc943SGeorge Wilson * Retrieve the vdev creation time. 61888ecc943SGeorge Wilson */ 61988ecc943SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_CREATE_TXG, 62088ecc943SGeorge Wilson &vd->vdev_crtxg); 62188ecc943SGeorge Wilson 622fa9e4066Sahrens /* 623fa9e4066Sahrens * If we're a top-level vdev, try to load the allocation parameters. 624fa9e4066Sahrens */ 6251195e687SMark J Musante if (parent && !parent->vdev_parent && 6261195e687SMark J Musante (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 627fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_ARRAY, 628fa9e4066Sahrens &vd->vdev_ms_array); 629fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_METASLAB_SHIFT, 630fa9e4066Sahrens &vd->vdev_ms_shift); 631fa9e4066Sahrens (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_ASIZE, 632fa9e4066Sahrens &vd->vdev_asize); 6333f9d6ad7SLin Ling (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVING, 6343f9d6ad7SLin Ling &vd->vdev_removing); 635215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_VDEV_TOP_ZAP, 636215198a6SJoe Stein &vd->vdev_top_zap); 637215198a6SJoe Stein } else { 638215198a6SJoe Stein ASSERT0(vd->vdev_top_zap); 639fa9e4066Sahrens } 640fa9e4066Sahrens 641cd0837ccSGeorge Wilson if (parent && !parent->vdev_parent && alloctype != VDEV_ALLOC_ATTACH) { 642a1521560SJeff Bonwick ASSERT(alloctype == VDEV_ALLOC_LOAD || 6439f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ADD || 6441195e687SMark J Musante alloctype == VDEV_ALLOC_SPLIT || 6459f4ab4d8SGeorge Wilson alloctype == VDEV_ALLOC_ROOTPOOL); 646a1521560SJeff Bonwick vd->vdev_mg = metaslab_group_create(islog ? 647a1521560SJeff Bonwick spa_log_class(spa) : spa_normal_class(spa), vd); 648a1521560SJeff Bonwick } 649a1521560SJeff Bonwick 650215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && 651215198a6SJoe Stein (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_SPLIT)) { 652215198a6SJoe Stein (void) nvlist_lookup_uint64(nv, 653215198a6SJoe Stein ZPOOL_CONFIG_VDEV_LEAF_ZAP, &vd->vdev_leaf_zap); 654215198a6SJoe Stein } else { 655215198a6SJoe Stein ASSERT0(vd->vdev_leaf_zap); 656215198a6SJoe Stein } 657215198a6SJoe Stein 658fa9e4066Sahrens /* 6593d7072f8Seschrock * If we're a leaf vdev, try to load the DTL object and other state. 660fa9e4066Sahrens */ 661215198a6SJoe Stein 662c5904d13Seschrock if (vd->vdev_ops->vdev_op_leaf && 66321ecdf64SLin Ling (alloctype == VDEV_ALLOC_LOAD || alloctype == VDEV_ALLOC_L2CACHE || 66421ecdf64SLin Ling alloctype == VDEV_ALLOC_ROOTPOOL)) { 665c5904d13Seschrock if (alloctype == VDEV_ALLOC_LOAD) { 666c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DTL, 6670713e232SGeorge Wilson &vd->vdev_dtl_object); 668c5904d13Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_UNSPARE, 669c5904d13Seschrock &vd->vdev_unspare); 670c5904d13Seschrock } 67121ecdf64SLin Ling 67221ecdf64SLin Ling if (alloctype == VDEV_ALLOC_ROOTPOOL) { 67321ecdf64SLin Ling uint64_t spare = 0; 67421ecdf64SLin Ling 67521ecdf64SLin Ling if (nvlist_lookup_uint64(nv, ZPOOL_CONFIG_IS_SPARE, 67621ecdf64SLin Ling &spare) == 0 && spare) 67721ecdf64SLin Ling spa_spare_add(vd); 67821ecdf64SLin Ling } 67921ecdf64SLin Ling 680ecc2d604Sbonwick (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_OFFLINE, 681ecc2d604Sbonwick &vd->vdev_offline); 682c5904d13Seschrock 683b4952e17SGeorge Wilson (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_RESILVER_TXG, 684b4952e17SGeorge Wilson &vd->vdev_resilver_txg); 685cb04b873SMark J Musante 6863d7072f8Seschrock /* 6873d7072f8Seschrock * When importing a pool, we want to ignore the persistent fault 6883d7072f8Seschrock * state, as the diagnosis made on another system may not be 689069f55e2SEric Schrock * valid in the current context. Local vdevs will 690069f55e2SEric Schrock * remain in the faulted state. 6913d7072f8Seschrock */ 692b16da2e2SGeorge Wilson if (spa_load_state(spa) == SPA_LOAD_OPEN) { 6933d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_FAULTED, 6943d7072f8Seschrock &vd->vdev_faulted); 6953d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_DEGRADED, 6963d7072f8Seschrock &vd->vdev_degraded); 6973d7072f8Seschrock (void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_REMOVED, 6983d7072f8Seschrock &vd->vdev_removed); 699069f55e2SEric Schrock 700069f55e2SEric Schrock if (vd->vdev_faulted || vd->vdev_degraded) { 701069f55e2SEric Schrock char *aux; 702069f55e2SEric Schrock 703069f55e2SEric Schrock vd->vdev_label_aux = 704069f55e2SEric Schrock VDEV_AUX_ERR_EXCEEDED; 705069f55e2SEric Schrock if (nvlist_lookup_string(nv, 706069f55e2SEric Schrock ZPOOL_CONFIG_AUX_STATE, &aux) == 0 && 707069f55e2SEric Schrock strcmp(aux, "external") == 0) 708069f55e2SEric Schrock vd->vdev_label_aux = VDEV_AUX_EXTERNAL; 709069f55e2SEric Schrock } 7103d7072f8Seschrock } 711fa9e4066Sahrens } 712fa9e4066Sahrens 713fa9e4066Sahrens /* 714fa9e4066Sahrens * Add ourselves to the parent's list of children. 715fa9e4066Sahrens */ 716fa9e4066Sahrens vdev_add_child(parent, vd); 717fa9e4066Sahrens 71899653d4eSeschrock *vdp = vd; 71999653d4eSeschrock 72099653d4eSeschrock return (0); 721fa9e4066Sahrens } 722fa9e4066Sahrens 723fa9e4066Sahrens void 724fa9e4066Sahrens vdev_free(vdev_t *vd) 725fa9e4066Sahrens { 7263d7072f8Seschrock spa_t *spa = vd->vdev_spa; 727fa9e4066Sahrens 728fa9e4066Sahrens /* 729fa9e4066Sahrens * vdev_free() implies closing the vdev first. This is simpler than 730fa9e4066Sahrens * trying to ensure complicated semantics for all callers. 731fa9e4066Sahrens */ 732fa9e4066Sahrens vdev_close(vd); 733fa9e4066Sahrens 734e14bb325SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_config_dirty_node)); 735b24ab676SJeff Bonwick ASSERT(!list_link_active(&vd->vdev_state_dirty_node)); 736fa9e4066Sahrens 737fa9e4066Sahrens /* 738fa9e4066Sahrens * Free all children. 739fa9e4066Sahrens */ 740573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 741fa9e4066Sahrens vdev_free(vd->vdev_child[c]); 742fa9e4066Sahrens 743fa9e4066Sahrens ASSERT(vd->vdev_child == NULL); 744fa9e4066Sahrens ASSERT(vd->vdev_guid_sum == vd->vdev_guid); 745fa9e4066Sahrens 746fa9e4066Sahrens /* 747fa9e4066Sahrens * Discard allocation state. 748fa9e4066Sahrens */ 749a1521560SJeff Bonwick if (vd->vdev_mg != NULL) { 750fa9e4066Sahrens vdev_metaslab_fini(vd); 751a1521560SJeff Bonwick metaslab_group_destroy(vd->vdev_mg); 752a1521560SJeff Bonwick } 753fa9e4066Sahrens 754fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_space); 755fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_dspace); 756fb09f5aaSMadhav Suresh ASSERT0(vd->vdev_stat.vs_alloc); 757fa9e4066Sahrens 758fa9e4066Sahrens /* 759fa9e4066Sahrens * Remove this vdev from its parent's child list. 760fa9e4066Sahrens */ 761fa9e4066Sahrens vdev_remove_child(vd->vdev_parent, vd); 762fa9e4066Sahrens 763fa9e4066Sahrens ASSERT(vd->vdev_parent == NULL); 764fa9e4066Sahrens 7653d7072f8Seschrock /* 7663d7072f8Seschrock * Clean up vdev structure. 7673d7072f8Seschrock */ 7683d7072f8Seschrock vdev_queue_fini(vd); 7693d7072f8Seschrock vdev_cache_fini(vd); 7703d7072f8Seschrock 7713d7072f8Seschrock if (vd->vdev_path) 7723d7072f8Seschrock spa_strfree(vd->vdev_path); 7733d7072f8Seschrock if (vd->vdev_devid) 7743d7072f8Seschrock spa_strfree(vd->vdev_devid); 7753d7072f8Seschrock if (vd->vdev_physpath) 7763d7072f8Seschrock spa_strfree(vd->vdev_physpath); 7776809eb4eSEric Schrock if (vd->vdev_fru) 7786809eb4eSEric Schrock spa_strfree(vd->vdev_fru); 7793d7072f8Seschrock 7803d7072f8Seschrock if (vd->vdev_isspare) 7813d7072f8Seschrock spa_spare_remove(vd); 782fa94a07fSbrendan if (vd->vdev_isl2cache) 783fa94a07fSbrendan spa_l2cache_remove(vd); 7843d7072f8Seschrock 7853d7072f8Seschrock txg_list_destroy(&vd->vdev_ms_list); 7863d7072f8Seschrock txg_list_destroy(&vd->vdev_dtl_list); 7878ad4d6ddSJeff Bonwick 7883d7072f8Seschrock mutex_enter(&vd->vdev_dtl_lock); 7890713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 7908ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 7910713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[t], NULL, NULL); 7920713e232SGeorge Wilson range_tree_destroy(vd->vdev_dtl[t]); 7938ad4d6ddSJeff Bonwick } 7943d7072f8Seschrock mutex_exit(&vd->vdev_dtl_lock); 7958ad4d6ddSJeff Bonwick 7965cabbc6bSPrashanth Sreenivasa EQUIV(vd->vdev_indirect_births != NULL, 7975cabbc6bSPrashanth Sreenivasa vd->vdev_indirect_mapping != NULL); 7985cabbc6bSPrashanth Sreenivasa if (vd->vdev_indirect_births != NULL) { 7995cabbc6bSPrashanth Sreenivasa vdev_indirect_mapping_close(vd->vdev_indirect_mapping); 8005cabbc6bSPrashanth Sreenivasa vdev_indirect_births_close(vd->vdev_indirect_births); 8015cabbc6bSPrashanth Sreenivasa } 8025cabbc6bSPrashanth Sreenivasa 8035cabbc6bSPrashanth Sreenivasa if (vd->vdev_obsolete_sm != NULL) { 8045cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_removing || 8055cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_indirect_ops); 8065cabbc6bSPrashanth Sreenivasa space_map_close(vd->vdev_obsolete_sm); 8075cabbc6bSPrashanth Sreenivasa vd->vdev_obsolete_sm = NULL; 8085cabbc6bSPrashanth Sreenivasa } 8095cabbc6bSPrashanth Sreenivasa range_tree_destroy(vd->vdev_obsolete_segments); 8105cabbc6bSPrashanth Sreenivasa rw_destroy(&vd->vdev_indirect_rwlock); 8115cabbc6bSPrashanth Sreenivasa mutex_destroy(&vd->vdev_obsolete_lock); 8125cabbc6bSPrashanth Sreenivasa 8130f7643c7SGeorge Wilson mutex_destroy(&vd->vdev_queue_lock); 8143d7072f8Seschrock mutex_destroy(&vd->vdev_dtl_lock); 8153d7072f8Seschrock mutex_destroy(&vd->vdev_stat_lock); 816e14bb325SJeff Bonwick mutex_destroy(&vd->vdev_probe_lock); 8173d7072f8Seschrock 8183d7072f8Seschrock if (vd == spa->spa_root_vdev) 8193d7072f8Seschrock spa->spa_root_vdev = NULL; 8203d7072f8Seschrock 8213d7072f8Seschrock kmem_free(vd, sizeof (vdev_t)); 822fa9e4066Sahrens } 823fa9e4066Sahrens 824fa9e4066Sahrens /* 825fa9e4066Sahrens * Transfer top-level vdev state from svd to tvd. 826fa9e4066Sahrens */ 827fa9e4066Sahrens static void 828fa9e4066Sahrens vdev_top_transfer(vdev_t *svd, vdev_t *tvd) 829fa9e4066Sahrens { 830fa9e4066Sahrens spa_t *spa = svd->vdev_spa; 831fa9e4066Sahrens metaslab_t *msp; 832fa9e4066Sahrens vdev_t *vd; 833fa9e4066Sahrens int t; 834fa9e4066Sahrens 835fa9e4066Sahrens ASSERT(tvd == tvd->vdev_top); 836fa9e4066Sahrens 837fa9e4066Sahrens tvd->vdev_ms_array = svd->vdev_ms_array; 838fa9e4066Sahrens tvd->vdev_ms_shift = svd->vdev_ms_shift; 839fa9e4066Sahrens tvd->vdev_ms_count = svd->vdev_ms_count; 840215198a6SJoe Stein tvd->vdev_top_zap = svd->vdev_top_zap; 841fa9e4066Sahrens 842fa9e4066Sahrens svd->vdev_ms_array = 0; 843fa9e4066Sahrens svd->vdev_ms_shift = 0; 844fa9e4066Sahrens svd->vdev_ms_count = 0; 845215198a6SJoe Stein svd->vdev_top_zap = 0; 846fa9e4066Sahrens 847cd0837ccSGeorge Wilson if (tvd->vdev_mg) 848cd0837ccSGeorge Wilson ASSERT3P(tvd->vdev_mg, ==, svd->vdev_mg); 849fa9e4066Sahrens tvd->vdev_mg = svd->vdev_mg; 850fa9e4066Sahrens tvd->vdev_ms = svd->vdev_ms; 851fa9e4066Sahrens 852fa9e4066Sahrens svd->vdev_mg = NULL; 853fa9e4066Sahrens svd->vdev_ms = NULL; 854ecc2d604Sbonwick 855ecc2d604Sbonwick if (tvd->vdev_mg != NULL) 856ecc2d604Sbonwick tvd->vdev_mg->mg_vd = tvd; 857fa9e4066Sahrens 858*86714001SSerapheim Dimitropoulos tvd->vdev_checkpoint_sm = svd->vdev_checkpoint_sm; 859*86714001SSerapheim Dimitropoulos svd->vdev_checkpoint_sm = NULL; 860*86714001SSerapheim Dimitropoulos 861fa9e4066Sahrens tvd->vdev_stat.vs_alloc = svd->vdev_stat.vs_alloc; 862fa9e4066Sahrens tvd->vdev_stat.vs_space = svd->vdev_stat.vs_space; 86399653d4eSeschrock tvd->vdev_stat.vs_dspace = svd->vdev_stat.vs_dspace; 864fa9e4066Sahrens 865fa9e4066Sahrens svd->vdev_stat.vs_alloc = 0; 866fa9e4066Sahrens svd->vdev_stat.vs_space = 0; 86799653d4eSeschrock svd->vdev_stat.vs_dspace = 0; 868fa9e4066Sahrens 869fa9e4066Sahrens for (t = 0; t < TXG_SIZE; t++) { 870fa9e4066Sahrens while ((msp = txg_list_remove(&svd->vdev_ms_list, t)) != NULL) 871fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_ms_list, msp, t); 872fa9e4066Sahrens while ((vd = txg_list_remove(&svd->vdev_dtl_list, t)) != NULL) 873fa9e4066Sahrens (void) txg_list_add(&tvd->vdev_dtl_list, vd, t); 874fa9e4066Sahrens if (txg_list_remove_this(&spa->spa_vdev_txg_list, svd, t)) 875fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, tvd, t); 876fa9e4066Sahrens } 877fa9e4066Sahrens 878e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_config_dirty_node)) { 879fa9e4066Sahrens vdev_config_clean(svd); 880fa9e4066Sahrens vdev_config_dirty(tvd); 881fa9e4066Sahrens } 882fa9e4066Sahrens 883e14bb325SJeff Bonwick if (list_link_active(&svd->vdev_state_dirty_node)) { 884e14bb325SJeff Bonwick vdev_state_clean(svd); 885e14bb325SJeff Bonwick vdev_state_dirty(tvd); 886e14bb325SJeff Bonwick } 887e14bb325SJeff Bonwick 88899653d4eSeschrock tvd->vdev_deflate_ratio = svd->vdev_deflate_ratio; 88999653d4eSeschrock svd->vdev_deflate_ratio = 0; 8908654d025Sperrin 8918654d025Sperrin tvd->vdev_islog = svd->vdev_islog; 8928654d025Sperrin svd->vdev_islog = 0; 893fa9e4066Sahrens } 894fa9e4066Sahrens 895fa9e4066Sahrens static void 896fa9e4066Sahrens vdev_top_update(vdev_t *tvd, vdev_t *vd) 897fa9e4066Sahrens { 898fa9e4066Sahrens if (vd == NULL) 899fa9e4066Sahrens return; 900fa9e4066Sahrens 901fa9e4066Sahrens vd->vdev_top = tvd; 902fa9e4066Sahrens 903573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 904fa9e4066Sahrens vdev_top_update(tvd, vd->vdev_child[c]); 905fa9e4066Sahrens } 906fa9e4066Sahrens 907fa9e4066Sahrens /* 908fa9e4066Sahrens * Add a mirror/replacing vdev above an existing vdev. 909fa9e4066Sahrens */ 910fa9e4066Sahrens vdev_t * 911fa9e4066Sahrens vdev_add_parent(vdev_t *cvd, vdev_ops_t *ops) 912fa9e4066Sahrens { 913fa9e4066Sahrens spa_t *spa = cvd->vdev_spa; 914fa9e4066Sahrens vdev_t *pvd = cvd->vdev_parent; 915fa9e4066Sahrens vdev_t *mvd; 916fa9e4066Sahrens 917e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_WRITER) == SCL_ALL); 918fa9e4066Sahrens 919fa9e4066Sahrens mvd = vdev_alloc_common(spa, cvd->vdev_id, 0, ops); 920ecc2d604Sbonwick 921ecc2d604Sbonwick mvd->vdev_asize = cvd->vdev_asize; 922573ca77eSGeorge Wilson mvd->vdev_min_asize = cvd->vdev_min_asize; 9234263d13fSGeorge Wilson mvd->vdev_max_asize = cvd->vdev_max_asize; 9245cabbc6bSPrashanth Sreenivasa mvd->vdev_psize = cvd->vdev_psize; 925ecc2d604Sbonwick mvd->vdev_ashift = cvd->vdev_ashift; 926ecc2d604Sbonwick mvd->vdev_state = cvd->vdev_state; 92788ecc943SGeorge Wilson mvd->vdev_crtxg = cvd->vdev_crtxg; 928ecc2d604Sbonwick 929fa9e4066Sahrens vdev_remove_child(pvd, cvd); 930fa9e4066Sahrens vdev_add_child(pvd, mvd); 931fa9e4066Sahrens cvd->vdev_id = mvd->vdev_children; 932fa9e4066Sahrens vdev_add_child(mvd, cvd); 933fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 934fa9e4066Sahrens 935fa9e4066Sahrens if (mvd == mvd->vdev_top) 936fa9e4066Sahrens vdev_top_transfer(cvd, mvd); 937fa9e4066Sahrens 938fa9e4066Sahrens return (mvd); 939fa9e4066Sahrens } 940fa9e4066Sahrens 941fa9e4066Sahrens /* 942fa9e4066Sahrens * Remove a 1-way mirror/replacing vdev from the tree. 943fa9e4066Sahrens */ 944fa9e4066Sahrens void 945fa9e4066Sahrens vdev_remove_parent(vdev_t *cvd) 946fa9e4066Sahrens { 947fa9e4066Sahrens vdev_t *mvd = cvd->vdev_parent; 948fa9e4066Sahrens vdev_t *pvd = mvd->vdev_parent; 949fa9e4066Sahrens 950e14bb325SJeff Bonwick ASSERT(spa_config_held(cvd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 951fa9e4066Sahrens 952fa9e4066Sahrens ASSERT(mvd->vdev_children == 1); 953fa9e4066Sahrens ASSERT(mvd->vdev_ops == &vdev_mirror_ops || 95499653d4eSeschrock mvd->vdev_ops == &vdev_replacing_ops || 95599653d4eSeschrock mvd->vdev_ops == &vdev_spare_ops); 956ecc2d604Sbonwick cvd->vdev_ashift = mvd->vdev_ashift; 957fa9e4066Sahrens 958fa9e4066Sahrens vdev_remove_child(mvd, cvd); 959fa9e4066Sahrens vdev_remove_child(pvd, mvd); 9608ad4d6ddSJeff Bonwick 96199653d4eSeschrock /* 962e14bb325SJeff Bonwick * If cvd will replace mvd as a top-level vdev, preserve mvd's guid. 963e14bb325SJeff Bonwick * Otherwise, we could have detached an offline device, and when we 964e14bb325SJeff Bonwick * go to import the pool we'll think we have two top-level vdevs, 965e14bb325SJeff Bonwick * instead of a different version of the same top-level vdev. 96699653d4eSeschrock */ 9678ad4d6ddSJeff Bonwick if (mvd->vdev_top == mvd) { 9688ad4d6ddSJeff Bonwick uint64_t guid_delta = mvd->vdev_guid - cvd->vdev_guid; 9691195e687SMark J Musante cvd->vdev_orig_guid = cvd->vdev_guid; 9708ad4d6ddSJeff Bonwick cvd->vdev_guid += guid_delta; 9718ad4d6ddSJeff Bonwick cvd->vdev_guid_sum += guid_delta; 9728ad4d6ddSJeff Bonwick } 973e14bb325SJeff Bonwick cvd->vdev_id = mvd->vdev_id; 974e14bb325SJeff Bonwick vdev_add_child(pvd, cvd); 975fa9e4066Sahrens vdev_top_update(cvd->vdev_top, cvd->vdev_top); 976fa9e4066Sahrens 977fa9e4066Sahrens if (cvd == cvd->vdev_top) 978fa9e4066Sahrens vdev_top_transfer(mvd, cvd); 979fa9e4066Sahrens 980fa9e4066Sahrens ASSERT(mvd->vdev_children == 0); 981fa9e4066Sahrens vdev_free(mvd); 982fa9e4066Sahrens } 983fa9e4066Sahrens 984ea8dc4b6Seschrock int 985fa9e4066Sahrens vdev_metaslab_init(vdev_t *vd, uint64_t txg) 986fa9e4066Sahrens { 987fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 988ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 989ecc2d604Sbonwick uint64_t m; 990fa9e4066Sahrens uint64_t oldc = vd->vdev_ms_count; 991fa9e4066Sahrens uint64_t newc = vd->vdev_asize >> vd->vdev_ms_shift; 992ecc2d604Sbonwick metaslab_t **mspp; 993ecc2d604Sbonwick int error; 994fa9e4066Sahrens 995a1521560SJeff Bonwick ASSERT(txg == 0 || spa_config_held(spa, SCL_ALLOC, RW_WRITER)); 996a1521560SJeff Bonwick 99788ecc943SGeorge Wilson /* 99888ecc943SGeorge Wilson * This vdev is not being allocated from yet or is a hole. 99988ecc943SGeorge Wilson */ 100088ecc943SGeorge Wilson if (vd->vdev_ms_shift == 0) 10010e34b6a7Sbonwick return (0); 10020e34b6a7Sbonwick 100388ecc943SGeorge Wilson ASSERT(!vd->vdev_ishole); 100488ecc943SGeorge Wilson 1005fa9e4066Sahrens ASSERT(oldc <= newc); 1006fa9e4066Sahrens 1007ecc2d604Sbonwick mspp = kmem_zalloc(newc * sizeof (*mspp), KM_SLEEP); 1008fa9e4066Sahrens 1009ecc2d604Sbonwick if (oldc != 0) { 1010ecc2d604Sbonwick bcopy(vd->vdev_ms, mspp, oldc * sizeof (*mspp)); 1011ecc2d604Sbonwick kmem_free(vd->vdev_ms, oldc * sizeof (*mspp)); 1012ecc2d604Sbonwick } 1013fa9e4066Sahrens 1014ecc2d604Sbonwick vd->vdev_ms = mspp; 1015ecc2d604Sbonwick vd->vdev_ms_count = newc; 1016fa9e4066Sahrens 1017ecc2d604Sbonwick for (m = oldc; m < newc; m++) { 10180713e232SGeorge Wilson uint64_t object = 0; 10190713e232SGeorge Wilson 10205cabbc6bSPrashanth Sreenivasa /* 10215cabbc6bSPrashanth Sreenivasa * vdev_ms_array may be 0 if we are creating the "fake" 10225cabbc6bSPrashanth Sreenivasa * metaslabs for an indirect vdev for zdb's leak detection. 10235cabbc6bSPrashanth Sreenivasa * See zdb_leak_init(). 10245cabbc6bSPrashanth Sreenivasa */ 10255cabbc6bSPrashanth Sreenivasa if (txg == 0 && vd->vdev_ms_array != 0) { 1026ecc2d604Sbonwick error = dmu_read(mos, vd->vdev_ms_array, 10277bfdf011SNeil Perrin m * sizeof (uint64_t), sizeof (uint64_t), &object, 10287bfdf011SNeil Perrin DMU_READ_PREFETCH); 10293ee8c80cSPavel Zakharov if (error != 0) { 10303ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "unable to read the metaslab " 10313ee8c80cSPavel Zakharov "array [error=%d]", error); 1032ecc2d604Sbonwick return (error); 10333ee8c80cSPavel Zakharov } 1034fa9e4066Sahrens } 10351e9bd7ecSPrakash Surya 10361e9bd7ecSPrakash Surya error = metaslab_init(vd->vdev_mg, m, object, txg, 10371e9bd7ecSPrakash Surya &(vd->vdev_ms[m])); 10383ee8c80cSPavel Zakharov if (error != 0) { 10393ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "metaslab_init failed [error=%d]", 10403ee8c80cSPavel Zakharov error); 10411e9bd7ecSPrakash Surya return (error); 10423ee8c80cSPavel Zakharov } 1043fa9e4066Sahrens } 1044fa9e4066Sahrens 1045a1521560SJeff Bonwick if (txg == 0) 1046a1521560SJeff Bonwick spa_config_enter(spa, SCL_ALLOC, FTAG, RW_WRITER); 1047a1521560SJeff Bonwick 10483f9d6ad7SLin Ling /* 10493f9d6ad7SLin Ling * If the vdev is being removed we don't activate 10503f9d6ad7SLin Ling * the metaslabs since we want to ensure that no new 10513f9d6ad7SLin Ling * allocations are performed on this device. 10523f9d6ad7SLin Ling */ 10533f9d6ad7SLin Ling if (oldc == 0 && !vd->vdev_removing) 1054a1521560SJeff Bonwick metaslab_group_activate(vd->vdev_mg); 1055a1521560SJeff Bonwick 1056a1521560SJeff Bonwick if (txg == 0) 1057a1521560SJeff Bonwick spa_config_exit(spa, SCL_ALLOC, FTAG); 1058a1521560SJeff Bonwick 1059ea8dc4b6Seschrock return (0); 1060fa9e4066Sahrens } 1061fa9e4066Sahrens 1062fa9e4066Sahrens void 1063fa9e4066Sahrens vdev_metaslab_fini(vdev_t *vd) 1064fa9e4066Sahrens { 1065*86714001SSerapheim Dimitropoulos if (vd->vdev_checkpoint_sm != NULL) { 1066*86714001SSerapheim Dimitropoulos ASSERT(spa_feature_is_active(vd->vdev_spa, 1067*86714001SSerapheim Dimitropoulos SPA_FEATURE_POOL_CHECKPOINT)); 1068*86714001SSerapheim Dimitropoulos space_map_close(vd->vdev_checkpoint_sm); 1069*86714001SSerapheim Dimitropoulos /* 1070*86714001SSerapheim Dimitropoulos * Even though we close the space map, we need to set its 1071*86714001SSerapheim Dimitropoulos * pointer to NULL. The reason is that vdev_metaslab_fini() 1072*86714001SSerapheim Dimitropoulos * may be called multiple times for certain operations 1073*86714001SSerapheim Dimitropoulos * (i.e. when destroying a pool) so we need to ensure that 1074*86714001SSerapheim Dimitropoulos * this clause never executes twice. This logic is similar 1075*86714001SSerapheim Dimitropoulos * to the one used for the vdev_ms clause below. 1076*86714001SSerapheim Dimitropoulos */ 1077*86714001SSerapheim Dimitropoulos vd->vdev_checkpoint_sm = NULL; 1078*86714001SSerapheim Dimitropoulos } 1079*86714001SSerapheim Dimitropoulos 1080fa9e4066Sahrens if (vd->vdev_ms != NULL) { 10815cabbc6bSPrashanth Sreenivasa uint64_t count = vd->vdev_ms_count; 10825cabbc6bSPrashanth Sreenivasa 1083a1521560SJeff Bonwick metaslab_group_passivate(vd->vdev_mg); 10845cabbc6bSPrashanth Sreenivasa for (uint64_t m = 0; m < count; m++) { 10850713e232SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 10860713e232SGeorge Wilson 10870713e232SGeorge Wilson if (msp != NULL) 10880713e232SGeorge Wilson metaslab_fini(msp); 10890713e232SGeorge Wilson } 1090fa9e4066Sahrens kmem_free(vd->vdev_ms, count * sizeof (metaslab_t *)); 1091fa9e4066Sahrens vd->vdev_ms = NULL; 10925cabbc6bSPrashanth Sreenivasa 10935cabbc6bSPrashanth Sreenivasa vd->vdev_ms_count = 0; 1094fa9e4066Sahrens } 10955cabbc6bSPrashanth Sreenivasa ASSERT0(vd->vdev_ms_count); 1096fa9e4066Sahrens } 1097fa9e4066Sahrens 1098e14bb325SJeff Bonwick typedef struct vdev_probe_stats { 1099e14bb325SJeff Bonwick boolean_t vps_readable; 1100e14bb325SJeff Bonwick boolean_t vps_writeable; 1101e14bb325SJeff Bonwick int vps_flags; 1102e14bb325SJeff Bonwick } vdev_probe_stats_t; 1103e14bb325SJeff Bonwick 1104e14bb325SJeff Bonwick static void 1105e14bb325SJeff Bonwick vdev_probe_done(zio_t *zio) 11060a4e9518Sgw { 11078ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 1108a3f829aeSBill Moore vdev_t *vd = zio->io_vd; 1109e14bb325SJeff Bonwick vdev_probe_stats_t *vps = zio->io_private; 1110a3f829aeSBill Moore 1111a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio != NULL); 1112e14bb325SJeff Bonwick 1113e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) { 1114e14bb325SJeff Bonwick if (zio->io_error == 0) 1115e14bb325SJeff Bonwick vps->vps_readable = 1; 11168ad4d6ddSJeff Bonwick if (zio->io_error == 0 && spa_writeable(spa)) { 1117a3f829aeSBill Moore zio_nowait(zio_write_phys(vd->vdev_probe_zio, vd, 1118770499e1SDan Kimmel zio->io_offset, zio->io_size, zio->io_abd, 1119e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1120e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_WRITE, vps->vps_flags, B_TRUE)); 1121e14bb325SJeff Bonwick } else { 1122770499e1SDan Kimmel abd_free(zio->io_abd); 1123e14bb325SJeff Bonwick } 1124e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_WRITE) { 1125e14bb325SJeff Bonwick if (zio->io_error == 0) 1126e14bb325SJeff Bonwick vps->vps_writeable = 1; 1127770499e1SDan Kimmel abd_free(zio->io_abd); 1128e14bb325SJeff Bonwick } else if (zio->io_type == ZIO_TYPE_NULL) { 1129a3f829aeSBill Moore zio_t *pio; 1130e14bb325SJeff Bonwick 1131e14bb325SJeff Bonwick vd->vdev_cant_read |= !vps->vps_readable; 1132e14bb325SJeff Bonwick vd->vdev_cant_write |= !vps->vps_writeable; 1133e14bb325SJeff Bonwick 1134e14bb325SJeff Bonwick if (vdev_readable(vd) && 11358ad4d6ddSJeff Bonwick (vdev_writeable(vd) || !spa_writeable(spa))) { 1136e14bb325SJeff Bonwick zio->io_error = 0; 1137e14bb325SJeff Bonwick } else { 1138e14bb325SJeff Bonwick ASSERT(zio->io_error != 0); 11393ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "failed probe"); 1140e14bb325SJeff Bonwick zfs_ereport_post(FM_EREPORT_ZFS_PROBE_FAILURE, 11418ad4d6ddSJeff Bonwick spa, vd, NULL, 0, 0); 1142be6fd75aSMatthew Ahrens zio->io_error = SET_ERROR(ENXIO); 1143e14bb325SJeff Bonwick } 1144a3f829aeSBill Moore 1145a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1146a3f829aeSBill Moore ASSERT(vd->vdev_probe_zio == zio); 1147a3f829aeSBill Moore vd->vdev_probe_zio = NULL; 1148a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1149a3f829aeSBill Moore 11500f7643c7SGeorge Wilson zio_link_t *zl = NULL; 11510f7643c7SGeorge Wilson while ((pio = zio_walk_parents(zio, &zl)) != NULL) 1152a3f829aeSBill Moore if (!vdev_accessible(vd, pio)) 1153be6fd75aSMatthew Ahrens pio->io_error = SET_ERROR(ENXIO); 1154a3f829aeSBill Moore 1155e14bb325SJeff Bonwick kmem_free(vps, sizeof (*vps)); 1156e14bb325SJeff Bonwick } 1157e14bb325SJeff Bonwick } 11580a4e9518Sgw 1159e14bb325SJeff Bonwick /* 1160f7170741SWill Andrews * Determine whether this device is accessible. 1161f7170741SWill Andrews * 1162f7170741SWill Andrews * Read and write to several known locations: the pad regions of each 1163f7170741SWill Andrews * vdev label but the first, which we leave alone in case it contains 1164f7170741SWill Andrews * a VTOC. 1165e14bb325SJeff Bonwick */ 1166e14bb325SJeff Bonwick zio_t * 1167a3f829aeSBill Moore vdev_probe(vdev_t *vd, zio_t *zio) 1168e14bb325SJeff Bonwick { 1169e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 1170a3f829aeSBill Moore vdev_probe_stats_t *vps = NULL; 1171a3f829aeSBill Moore zio_t *pio; 1172a3f829aeSBill Moore 1173a3f829aeSBill Moore ASSERT(vd->vdev_ops->vdev_op_leaf); 11740a4e9518Sgw 1175a3f829aeSBill Moore /* 1176a3f829aeSBill Moore * Don't probe the probe. 1177a3f829aeSBill Moore */ 1178a3f829aeSBill Moore if (zio && (zio->io_flags & ZIO_FLAG_PROBE)) 1179a3f829aeSBill Moore return (NULL); 1180e14bb325SJeff Bonwick 1181a3f829aeSBill Moore /* 1182a3f829aeSBill Moore * To prevent 'probe storms' when a device fails, we create 1183a3f829aeSBill Moore * just one probe i/o at a time. All zios that want to probe 1184a3f829aeSBill Moore * this vdev will become parents of the probe io. 1185a3f829aeSBill Moore */ 1186a3f829aeSBill Moore mutex_enter(&vd->vdev_probe_lock); 1187e14bb325SJeff Bonwick 1188a3f829aeSBill Moore if ((pio = vd->vdev_probe_zio) == NULL) { 1189a3f829aeSBill Moore vps = kmem_zalloc(sizeof (*vps), KM_SLEEP); 1190a3f829aeSBill Moore 1191a3f829aeSBill Moore vps->vps_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_PROBE | 1192a3f829aeSBill Moore ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE | 11938956713aSEric Schrock ZIO_FLAG_TRYHARD; 1194a3f829aeSBill Moore 1195a3f829aeSBill Moore if (spa_config_held(spa, SCL_ZIO, RW_WRITER)) { 1196a3f829aeSBill Moore /* 1197a3f829aeSBill Moore * vdev_cant_read and vdev_cant_write can only 1198a3f829aeSBill Moore * transition from TRUE to FALSE when we have the 1199a3f829aeSBill Moore * SCL_ZIO lock as writer; otherwise they can only 1200a3f829aeSBill Moore * transition from FALSE to TRUE. This ensures that 1201a3f829aeSBill Moore * any zio looking at these values can assume that 1202a3f829aeSBill Moore * failures persist for the life of the I/O. That's 1203a3f829aeSBill Moore * important because when a device has intermittent 1204a3f829aeSBill Moore * connectivity problems, we want to ensure that 1205a3f829aeSBill Moore * they're ascribed to the device (ENXIO) and not 1206a3f829aeSBill Moore * the zio (EIO). 1207a3f829aeSBill Moore * 1208a3f829aeSBill Moore * Since we hold SCL_ZIO as writer here, clear both 1209a3f829aeSBill Moore * values so the probe can reevaluate from first 1210a3f829aeSBill Moore * principles. 1211a3f829aeSBill Moore */ 1212a3f829aeSBill Moore vps->vps_flags |= ZIO_FLAG_CONFIG_WRITER; 1213a3f829aeSBill Moore vd->vdev_cant_read = B_FALSE; 1214a3f829aeSBill Moore vd->vdev_cant_write = B_FALSE; 1215a3f829aeSBill Moore } 1216a3f829aeSBill Moore 1217a3f829aeSBill Moore vd->vdev_probe_zio = pio = zio_null(NULL, spa, vd, 1218a3f829aeSBill Moore vdev_probe_done, vps, 1219a3f829aeSBill Moore vps->vps_flags | ZIO_FLAG_DONT_PROPAGATE); 1220a3f829aeSBill Moore 122198d1cbfeSGeorge Wilson /* 122298d1cbfeSGeorge Wilson * We can't change the vdev state in this context, so we 122398d1cbfeSGeorge Wilson * kick off an async task to do it on our behalf. 122498d1cbfeSGeorge Wilson */ 1225a3f829aeSBill Moore if (zio != NULL) { 1226a3f829aeSBill Moore vd->vdev_probe_wanted = B_TRUE; 1227a3f829aeSBill Moore spa_async_request(spa, SPA_ASYNC_PROBE); 1228a3f829aeSBill Moore } 1229e14bb325SJeff Bonwick } 1230e14bb325SJeff Bonwick 1231a3f829aeSBill Moore if (zio != NULL) 1232a3f829aeSBill Moore zio_add_child(zio, pio); 1233e14bb325SJeff Bonwick 1234a3f829aeSBill Moore mutex_exit(&vd->vdev_probe_lock); 1235e14bb325SJeff Bonwick 1236a3f829aeSBill Moore if (vps == NULL) { 1237a3f829aeSBill Moore ASSERT(zio != NULL); 1238a3f829aeSBill Moore return (NULL); 1239a3f829aeSBill Moore } 1240e14bb325SJeff Bonwick 1241e14bb325SJeff Bonwick for (int l = 1; l < VDEV_LABELS; l++) { 1242a3f829aeSBill Moore zio_nowait(zio_read_phys(pio, vd, 1243e14bb325SJeff Bonwick vdev_label_offset(vd->vdev_psize, l, 1244770499e1SDan Kimmel offsetof(vdev_label_t, vl_pad2)), VDEV_PAD_SIZE, 1245770499e1SDan Kimmel abd_alloc_for_io(VDEV_PAD_SIZE, B_TRUE), 1246e14bb325SJeff Bonwick ZIO_CHECKSUM_OFF, vdev_probe_done, vps, 1247e14bb325SJeff Bonwick ZIO_PRIORITY_SYNC_READ, vps->vps_flags, B_TRUE)); 1248e14bb325SJeff Bonwick } 1249e14bb325SJeff Bonwick 1250a3f829aeSBill Moore if (zio == NULL) 1251a3f829aeSBill Moore return (pio); 1252a3f829aeSBill Moore 1253a3f829aeSBill Moore zio_nowait(pio); 1254a3f829aeSBill Moore return (NULL); 12550a4e9518Sgw } 12560a4e9518Sgw 1257f64c0e34SEric Taylor static void 1258f64c0e34SEric Taylor vdev_open_child(void *arg) 1259f64c0e34SEric Taylor { 1260f64c0e34SEric Taylor vdev_t *vd = arg; 1261f64c0e34SEric Taylor 1262f64c0e34SEric Taylor vd->vdev_open_thread = curthread; 1263f64c0e34SEric Taylor vd->vdev_open_error = vdev_open(vd); 1264f64c0e34SEric Taylor vd->vdev_open_thread = NULL; 1265f64c0e34SEric Taylor } 1266f64c0e34SEric Taylor 1267681d9761SEric Taylor boolean_t 1268681d9761SEric Taylor vdev_uses_zvols(vdev_t *vd) 1269681d9761SEric Taylor { 1270681d9761SEric Taylor if (vd->vdev_path && strncmp(vd->vdev_path, ZVOL_DIR, 1271681d9761SEric Taylor strlen(ZVOL_DIR)) == 0) 1272681d9761SEric Taylor return (B_TRUE); 1273681d9761SEric Taylor for (int c = 0; c < vd->vdev_children; c++) 1274681d9761SEric Taylor if (vdev_uses_zvols(vd->vdev_child[c])) 1275681d9761SEric Taylor return (B_TRUE); 1276681d9761SEric Taylor return (B_FALSE); 1277681d9761SEric Taylor } 1278681d9761SEric Taylor 1279f64c0e34SEric Taylor void 1280f64c0e34SEric Taylor vdev_open_children(vdev_t *vd) 1281f64c0e34SEric Taylor { 1282f64c0e34SEric Taylor taskq_t *tq; 1283f64c0e34SEric Taylor int children = vd->vdev_children; 1284f64c0e34SEric Taylor 1285681d9761SEric Taylor /* 1286681d9761SEric Taylor * in order to handle pools on top of zvols, do the opens 1287681d9761SEric Taylor * in a single thread so that the same thread holds the 1288681d9761SEric Taylor * spa_namespace_lock 1289681d9761SEric Taylor */ 1290681d9761SEric Taylor if (vdev_uses_zvols(vd)) { 1291681d9761SEric Taylor for (int c = 0; c < children; c++) 1292681d9761SEric Taylor vd->vdev_child[c]->vdev_open_error = 1293681d9761SEric Taylor vdev_open(vd->vdev_child[c]); 1294681d9761SEric Taylor return; 1295681d9761SEric Taylor } 1296f64c0e34SEric Taylor tq = taskq_create("vdev_open", children, minclsyspri, 1297f64c0e34SEric Taylor children, children, TASKQ_PREPOPULATE); 1298f64c0e34SEric Taylor 1299f64c0e34SEric Taylor for (int c = 0; c < children; c++) 1300f64c0e34SEric Taylor VERIFY(taskq_dispatch(tq, vdev_open_child, vd->vdev_child[c], 1301f64c0e34SEric Taylor TQ_SLEEP) != NULL); 1302f64c0e34SEric Taylor 1303f64c0e34SEric Taylor taskq_destroy(tq); 1304f64c0e34SEric Taylor } 1305f64c0e34SEric Taylor 13065cabbc6bSPrashanth Sreenivasa /* 13075cabbc6bSPrashanth Sreenivasa * Compute the raidz-deflation ratio. Note, we hard-code 13085cabbc6bSPrashanth Sreenivasa * in 128k (1 << 17) because it is the "typical" blocksize. 13095cabbc6bSPrashanth Sreenivasa * Even though SPA_MAXBLOCKSIZE changed, this algorithm can not change, 13105cabbc6bSPrashanth Sreenivasa * otherwise it would inconsistently account for existing bp's. 13115cabbc6bSPrashanth Sreenivasa */ 13125cabbc6bSPrashanth Sreenivasa static void 13135cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vdev_t *vd) 13145cabbc6bSPrashanth Sreenivasa { 13155cabbc6bSPrashanth Sreenivasa if (vd == vd->vdev_top && !vd->vdev_ishole && vd->vdev_ashift != 0) { 13165cabbc6bSPrashanth Sreenivasa vd->vdev_deflate_ratio = (1 << 17) / 13175cabbc6bSPrashanth Sreenivasa (vdev_psize_to_asize(vd, 1 << 17) >> SPA_MINBLOCKSHIFT); 13185cabbc6bSPrashanth Sreenivasa } 13195cabbc6bSPrashanth Sreenivasa } 13205cabbc6bSPrashanth Sreenivasa 1321fa9e4066Sahrens /* 1322fa9e4066Sahrens * Prepare a virtual device for access. 1323fa9e4066Sahrens */ 1324fa9e4066Sahrens int 1325fa9e4066Sahrens vdev_open(vdev_t *vd) 1326fa9e4066Sahrens { 13278ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1328fa9e4066Sahrens int error; 1329fa9e4066Sahrens uint64_t osize = 0; 13304263d13fSGeorge Wilson uint64_t max_osize = 0; 13314263d13fSGeorge Wilson uint64_t asize, max_asize, psize; 1332ecc2d604Sbonwick uint64_t ashift = 0; 1333fa9e4066Sahrens 1334f64c0e34SEric Taylor ASSERT(vd->vdev_open_thread == curthread || 1335f64c0e34SEric Taylor spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1336fa9e4066Sahrens ASSERT(vd->vdev_state == VDEV_STATE_CLOSED || 1337fa9e4066Sahrens vd->vdev_state == VDEV_STATE_CANT_OPEN || 1338fa9e4066Sahrens vd->vdev_state == VDEV_STATE_OFFLINE); 1339fa9e4066Sahrens 1340fa9e4066Sahrens vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1341e6ca193dSGeorge Wilson vd->vdev_cant_read = B_FALSE; 1342e6ca193dSGeorge Wilson vd->vdev_cant_write = B_FALSE; 1343573ca77eSGeorge Wilson vd->vdev_min_asize = vdev_get_min_asize(vd); 1344fa9e4066Sahrens 1345069f55e2SEric Schrock /* 1346069f55e2SEric Schrock * If this vdev is not removed, check its fault status. If it's 1347069f55e2SEric Schrock * faulted, bail out of the open. 1348069f55e2SEric Schrock */ 13493d7072f8Seschrock if (!vd->vdev_removed && vd->vdev_faulted) { 13503d7072f8Seschrock ASSERT(vd->vdev_children == 0); 1351069f55e2SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1352069f55e2SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 13533d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1354069f55e2SEric Schrock vd->vdev_label_aux); 1355be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 13563d7072f8Seschrock } else if (vd->vdev_offline) { 1357fa9e4066Sahrens ASSERT(vd->vdev_children == 0); 1358ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, VDEV_AUX_NONE); 1359be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1360fa9e4066Sahrens } 1361fa9e4066Sahrens 13624263d13fSGeorge Wilson error = vd->vdev_ops->vdev_op_open(vd, &osize, &max_osize, &ashift); 1363fa9e4066Sahrens 1364095bcd66SGeorge Wilson /* 1365095bcd66SGeorge Wilson * Reset the vdev_reopening flag so that we actually close 1366095bcd66SGeorge Wilson * the vdev on error. 1367095bcd66SGeorge Wilson */ 1368095bcd66SGeorge Wilson vd->vdev_reopening = B_FALSE; 1369ea8dc4b6Seschrock if (zio_injection_enabled && error == 0) 13708956713aSEric Schrock error = zio_handle_device_injection(vd, NULL, ENXIO); 1371ea8dc4b6Seschrock 1372fa9e4066Sahrens if (error) { 13733d7072f8Seschrock if (vd->vdev_removed && 13743d7072f8Seschrock vd->vdev_stat.vs_aux != VDEV_AUX_OPEN_FAILED) 13753d7072f8Seschrock vd->vdev_removed = B_FALSE; 13763d7072f8Seschrock 13776f793812SPavel Zakharov if (vd->vdev_stat.vs_aux == VDEV_AUX_CHILDREN_OFFLINE) { 13786f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_OFFLINE, 13796f793812SPavel Zakharov vd->vdev_stat.vs_aux); 13806f793812SPavel Zakharov } else { 13816f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 13826f793812SPavel Zakharov vd->vdev_stat.vs_aux); 13836f793812SPavel Zakharov } 1384fa9e4066Sahrens return (error); 1385fa9e4066Sahrens } 1386fa9e4066Sahrens 13873d7072f8Seschrock vd->vdev_removed = B_FALSE; 13883d7072f8Seschrock 1389096d22d4SEric Schrock /* 1390096d22d4SEric Schrock * Recheck the faulted flag now that we have confirmed that 1391096d22d4SEric Schrock * the vdev is accessible. If we're faulted, bail. 1392096d22d4SEric Schrock */ 1393096d22d4SEric Schrock if (vd->vdev_faulted) { 1394096d22d4SEric Schrock ASSERT(vd->vdev_children == 0); 1395096d22d4SEric Schrock ASSERT(vd->vdev_label_aux == VDEV_AUX_ERR_EXCEEDED || 1396096d22d4SEric Schrock vd->vdev_label_aux == VDEV_AUX_EXTERNAL); 1397096d22d4SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 1398096d22d4SEric Schrock vd->vdev_label_aux); 1399be6fd75aSMatthew Ahrens return (SET_ERROR(ENXIO)); 1400096d22d4SEric Schrock } 1401096d22d4SEric Schrock 14023d7072f8Seschrock if (vd->vdev_degraded) { 14033d7072f8Seschrock ASSERT(vd->vdev_children == 0); 14043d7072f8Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 14053d7072f8Seschrock VDEV_AUX_ERR_EXCEEDED); 14063d7072f8Seschrock } else { 1407069f55e2SEric Schrock vdev_set_state(vd, B_TRUE, VDEV_STATE_HEALTHY, 0); 14083d7072f8Seschrock } 1409fa9e4066Sahrens 141088ecc943SGeorge Wilson /* 141188ecc943SGeorge Wilson * For hole or missing vdevs we just return success. 141288ecc943SGeorge Wilson */ 141388ecc943SGeorge Wilson if (vd->vdev_ishole || vd->vdev_ops == &vdev_missing_ops) 141488ecc943SGeorge Wilson return (0); 141588ecc943SGeorge Wilson 1416573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 1417ea8dc4b6Seschrock if (vd->vdev_child[c]->vdev_state != VDEV_STATE_HEALTHY) { 1418ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_DEGRADED, 1419ea8dc4b6Seschrock VDEV_AUX_NONE); 1420ea8dc4b6Seschrock break; 1421ea8dc4b6Seschrock } 1422573ca77eSGeorge Wilson } 1423fa9e4066Sahrens 1424fa9e4066Sahrens osize = P2ALIGN(osize, (uint64_t)sizeof (vdev_label_t)); 14254263d13fSGeorge Wilson max_osize = P2ALIGN(max_osize, (uint64_t)sizeof (vdev_label_t)); 1426fa9e4066Sahrens 1427fa9e4066Sahrens if (vd->vdev_children == 0) { 1428fa9e4066Sahrens if (osize < SPA_MINDEVSIZE) { 1429ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1430ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1431be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1432fa9e4066Sahrens } 1433fa9e4066Sahrens psize = osize; 1434fa9e4066Sahrens asize = osize - (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE); 14354263d13fSGeorge Wilson max_asize = max_osize - (VDEV_LABEL_START_SIZE + 14364263d13fSGeorge Wilson VDEV_LABEL_END_SIZE); 1437fa9e4066Sahrens } else { 1438ecc2d604Sbonwick if (vd->vdev_parent != NULL && osize < SPA_MINDEVSIZE - 1439fa9e4066Sahrens (VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE)) { 1440ea8dc4b6Seschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1441ea8dc4b6Seschrock VDEV_AUX_TOO_SMALL); 1442be6fd75aSMatthew Ahrens return (SET_ERROR(EOVERFLOW)); 1443fa9e4066Sahrens } 1444fa9e4066Sahrens psize = 0; 1445fa9e4066Sahrens asize = osize; 14464263d13fSGeorge Wilson max_asize = max_osize; 1447fa9e4066Sahrens } 1448fa9e4066Sahrens 1449fa9e4066Sahrens vd->vdev_psize = psize; 1450fa9e4066Sahrens 1451573ca77eSGeorge Wilson /* 1452c040c10cSSteven Hartland * Make sure the allocatable size hasn't shrunk too much. 1453573ca77eSGeorge Wilson */ 1454573ca77eSGeorge Wilson if (asize < vd->vdev_min_asize) { 1455573ca77eSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 1456573ca77eSGeorge Wilson VDEV_AUX_BAD_LABEL); 1457be6fd75aSMatthew Ahrens return (SET_ERROR(EINVAL)); 1458573ca77eSGeorge Wilson } 1459573ca77eSGeorge Wilson 1460fa9e4066Sahrens if (vd->vdev_asize == 0) { 1461fa9e4066Sahrens /* 1462fa9e4066Sahrens * This is the first-ever open, so use the computed values. 1463ecc2d604Sbonwick * For testing purposes, a higher ashift can be requested. 1464fa9e4066Sahrens */ 1465fa9e4066Sahrens vd->vdev_asize = asize; 14664263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1467ecc2d604Sbonwick vd->vdev_ashift = MAX(ashift, vd->vdev_ashift); 1468fa9e4066Sahrens } else { 1469fa9e4066Sahrens /* 14702384d9f8SGeorge Wilson * Detect if the alignment requirement has increased. 14712384d9f8SGeorge Wilson * We don't want to make the pool unavailable, just 14722384d9f8SGeorge Wilson * issue a warning instead. 1473fa9e4066Sahrens */ 14742384d9f8SGeorge Wilson if (ashift > vd->vdev_top->vdev_ashift && 14752384d9f8SGeorge Wilson vd->vdev_ops->vdev_op_leaf) { 14762384d9f8SGeorge Wilson cmn_err(CE_WARN, 14772384d9f8SGeorge Wilson "Disk, '%s', has a block alignment that is " 14782384d9f8SGeorge Wilson "larger than the pool's alignment\n", 14792384d9f8SGeorge Wilson vd->vdev_path); 1480fa9e4066Sahrens } 14814263d13fSGeorge Wilson vd->vdev_max_asize = max_asize; 1482573ca77eSGeorge Wilson } 1483fa9e4066Sahrens 1484573ca77eSGeorge Wilson /* 1485c040c10cSSteven Hartland * If all children are healthy we update asize if either: 1486c040c10cSSteven Hartland * The asize has increased, due to a device expansion caused by dynamic 1487c040c10cSSteven Hartland * LUN growth or vdev replacement, and automatic expansion is enabled; 1488c040c10cSSteven Hartland * making the additional space available. 1489c040c10cSSteven Hartland * 1490c040c10cSSteven Hartland * The asize has decreased, due to a device shrink usually caused by a 1491c040c10cSSteven Hartland * vdev replace with a smaller device. This ensures that calculations 1492c040c10cSSteven Hartland * based of max_asize and asize e.g. esize are always valid. It's safe 1493c040c10cSSteven Hartland * to do this as we've already validated that asize is greater than 1494c040c10cSSteven Hartland * vdev_min_asize. 1495573ca77eSGeorge Wilson */ 1496c040c10cSSteven Hartland if (vd->vdev_state == VDEV_STATE_HEALTHY && 1497c040c10cSSteven Hartland ((asize > vd->vdev_asize && 1498c040c10cSSteven Hartland (vd->vdev_expanding || spa->spa_autoexpand)) || 1499c040c10cSSteven Hartland (asize < vd->vdev_asize))) 1500573ca77eSGeorge Wilson vd->vdev_asize = asize; 1501fa9e4066Sahrens 1502573ca77eSGeorge Wilson vdev_set_min_asize(vd); 1503fa9e4066Sahrens 15040a4e9518Sgw /* 15050a4e9518Sgw * Ensure we can issue some IO before declaring the 15060a4e9518Sgw * vdev open for business. 15070a4e9518Sgw */ 1508e14bb325SJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && 1509e14bb325SJeff Bonwick (error = zio_wait(vdev_probe(vd, NULL))) != 0) { 151098d1cbfeSGeorge Wilson vdev_set_state(vd, B_TRUE, VDEV_STATE_FAULTED, 151198d1cbfeSGeorge Wilson VDEV_AUX_ERR_EXCEEDED); 15120a4e9518Sgw return (error); 15130a4e9518Sgw } 15140a4e9518Sgw 151581cd5c55SMatthew Ahrens /* 151681cd5c55SMatthew Ahrens * Track the min and max ashift values for normal data devices. 151781cd5c55SMatthew Ahrens */ 151881cd5c55SMatthew Ahrens if (vd->vdev_top == vd && vd->vdev_ashift != 0 && 151981cd5c55SMatthew Ahrens !vd->vdev_islog && vd->vdev_aux == NULL) { 152081cd5c55SMatthew Ahrens if (vd->vdev_ashift > spa->spa_max_ashift) 152181cd5c55SMatthew Ahrens spa->spa_max_ashift = vd->vdev_ashift; 152281cd5c55SMatthew Ahrens if (vd->vdev_ashift < spa->spa_min_ashift) 152381cd5c55SMatthew Ahrens spa->spa_min_ashift = vd->vdev_ashift; 152481cd5c55SMatthew Ahrens } 152581cd5c55SMatthew Ahrens 1526088f3894Sahrens /* 1527088f3894Sahrens * If a leaf vdev has a DTL, and seems healthy, then kick off a 15288ad4d6ddSJeff Bonwick * resilver. But don't do this if we are doing a reopen for a scrub, 15298ad4d6ddSJeff Bonwick * since this would just restart the scrub we are already doing. 1530088f3894Sahrens */ 15318ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf && !spa->spa_scrub_reopen && 15328ad4d6ddSJeff Bonwick vdev_resilver_needed(vd, NULL, NULL)) 15338ad4d6ddSJeff Bonwick spa_async_request(spa, SPA_ASYNC_RESILVER); 1534088f3894Sahrens 1535fa9e4066Sahrens return (0); 1536fa9e4066Sahrens } 1537fa9e4066Sahrens 1538560e6e96Seschrock /* 1539560e6e96Seschrock * Called once the vdevs are all opened, this routine validates the label 15406f793812SPavel Zakharov * contents. This needs to be done before vdev_load() so that we don't 15413d7072f8Seschrock * inadvertently do repair I/Os to the wrong device. 1542560e6e96Seschrock * 1543560e6e96Seschrock * This function will only return failure if one of the vdevs indicates that it 1544560e6e96Seschrock * has since been destroyed or exported. This is only possible if 1545560e6e96Seschrock * /etc/zfs/zpool.cache was readonly at the time. Otherwise, the vdev state 1546560e6e96Seschrock * will be updated but the function will return 0. 1547560e6e96Seschrock */ 1548560e6e96Seschrock int 15496f793812SPavel Zakharov vdev_validate(vdev_t *vd) 1550560e6e96Seschrock { 1551560e6e96Seschrock spa_t *spa = vd->vdev_spa; 1552560e6e96Seschrock nvlist_t *label; 15536f793812SPavel Zakharov uint64_t guid = 0, aux_guid = 0, top_guid; 1554560e6e96Seschrock uint64_t state; 15556f793812SPavel Zakharov nvlist_t *nvl; 15566f793812SPavel Zakharov uint64_t txg; 1557560e6e96Seschrock 15586f793812SPavel Zakharov if (vdev_validate_skip) 15596f793812SPavel Zakharov return (0); 15606f793812SPavel Zakharov 15616f793812SPavel Zakharov for (uint64_t c = 0; c < vd->vdev_children; c++) 15626f793812SPavel Zakharov if (vdev_validate(vd->vdev_child[c]) != 0) 1563be6fd75aSMatthew Ahrens return (SET_ERROR(EBADF)); 1564560e6e96Seschrock 1565b5989ec7Seschrock /* 1566b5989ec7Seschrock * If the device has already failed, or was marked offline, don't do 1567b5989ec7Seschrock * any further validation. Otherwise, label I/O will fail and we will 1568b5989ec7Seschrock * overwrite the previous state. 1569b5989ec7Seschrock */ 15706f793812SPavel Zakharov if (!vd->vdev_ops->vdev_op_leaf || !vdev_readable(vd)) 15716f793812SPavel Zakharov return (0); 1572560e6e96Seschrock 15736f793812SPavel Zakharov /* 15746f793812SPavel Zakharov * If we are performing an extreme rewind, we allow for a label that 15756f793812SPavel Zakharov * was modified at a point after the current txg. 15766f793812SPavel Zakharov */ 15776f793812SPavel Zakharov if (spa->spa_extreme_rewind || spa_last_synced_txg(spa) == 0) 15786f793812SPavel Zakharov txg = UINT64_MAX; 15796f793812SPavel Zakharov else 15806f793812SPavel Zakharov txg = spa_last_synced_txg(spa); 1581560e6e96Seschrock 15826f793812SPavel Zakharov if ((label = vdev_label_read_config(vd, txg)) == NULL) { 15836f793812SPavel Zakharov vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 15846f793812SPavel Zakharov VDEV_AUX_BAD_LABEL); 15856f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: failed reading config"); 15866f793812SPavel Zakharov return (0); 15876f793812SPavel Zakharov } 15881195e687SMark J Musante 15896f793812SPavel Zakharov /* 15906f793812SPavel Zakharov * Determine if this vdev has been split off into another 15916f793812SPavel Zakharov * pool. If so, then refuse to open it. 15926f793812SPavel Zakharov */ 15936f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_SPLIT_GUID, 15946f793812SPavel Zakharov &aux_guid) == 0 && aux_guid == spa_guid(spa)) { 15956f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 15966f793812SPavel Zakharov VDEV_AUX_SPLIT_POOL); 15976f793812SPavel Zakharov nvlist_free(label); 15986f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: vdev split into other pool"); 15996f793812SPavel Zakharov return (0); 16006f793812SPavel Zakharov } 1601560e6e96Seschrock 16026f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_GUID, &guid) != 0) { 16036f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 16046f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 16056f793812SPavel Zakharov nvlist_free(label); 16066f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 16076f793812SPavel Zakharov ZPOOL_CONFIG_POOL_GUID); 16086f793812SPavel Zakharov return (0); 16096f793812SPavel Zakharov } 16101195e687SMark J Musante 16116f793812SPavel Zakharov /* 16126f793812SPavel Zakharov * If config is not trusted then ignore the spa guid check. This is 16136f793812SPavel Zakharov * necessary because if the machine crashed during a re-guid the new 16146f793812SPavel Zakharov * guid might have been written to all of the vdev labels, but not the 16156f793812SPavel Zakharov * cached config. The check will be performed again once we have the 16166f793812SPavel Zakharov * trusted config from the MOS. 16176f793812SPavel Zakharov */ 16186f793812SPavel Zakharov if (spa->spa_trust_config && guid != spa_guid(spa)) { 16196f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 16206f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 16216f793812SPavel Zakharov nvlist_free(label); 16226f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: vdev label pool_guid doesn't " 16236f793812SPavel Zakharov "match config (%llu != %llu)", (u_longlong_t)guid, 16246f793812SPavel Zakharov (u_longlong_t)spa_guid(spa)); 16256f793812SPavel Zakharov return (0); 16266f793812SPavel Zakharov } 16276f793812SPavel Zakharov 16286f793812SPavel Zakharov if (nvlist_lookup_nvlist(label, ZPOOL_CONFIG_VDEV_TREE, &nvl) 16296f793812SPavel Zakharov != 0 || nvlist_lookup_uint64(nvl, ZPOOL_CONFIG_ORIG_GUID, 16306f793812SPavel Zakharov &aux_guid) != 0) 16316f793812SPavel Zakharov aux_guid = 0; 16326f793812SPavel Zakharov 16336f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0) { 16346f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 16356f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 16366f793812SPavel Zakharov nvlist_free(label); 16376f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 16386f793812SPavel Zakharov ZPOOL_CONFIG_GUID); 16396f793812SPavel Zakharov return (0); 16406f793812SPavel Zakharov } 16416f793812SPavel Zakharov 16426f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_TOP_GUID, &top_guid) 16436f793812SPavel Zakharov != 0) { 16446f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 16456f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 16466f793812SPavel Zakharov nvlist_free(label); 16476f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 16486f793812SPavel Zakharov ZPOOL_CONFIG_TOP_GUID); 16496f793812SPavel Zakharov return (0); 16506f793812SPavel Zakharov } 16516f793812SPavel Zakharov 16526f793812SPavel Zakharov /* 16536f793812SPavel Zakharov * If this vdev just became a top-level vdev because its sibling was 16546f793812SPavel Zakharov * detached, it will have adopted the parent's vdev guid -- but the 16556f793812SPavel Zakharov * label may or may not be on disk yet. Fortunately, either version 16566f793812SPavel Zakharov * of the label will have the same top guid, so if we're a top-level 16576f793812SPavel Zakharov * vdev, we can safely compare to that instead. 16586f793812SPavel Zakharov * However, if the config comes from a cachefile that failed to update 16596f793812SPavel Zakharov * after the detach, a top-level vdev will appear as a non top-level 16606f793812SPavel Zakharov * vdev in the config. Also relax the constraints if we perform an 16616f793812SPavel Zakharov * extreme rewind. 16626f793812SPavel Zakharov * 16636f793812SPavel Zakharov * If we split this vdev off instead, then we also check the 16646f793812SPavel Zakharov * original pool's guid. We don't want to consider the vdev 16656f793812SPavel Zakharov * corrupt if it is partway through a split operation. 16666f793812SPavel Zakharov */ 16676f793812SPavel Zakharov if (vd->vdev_guid != guid && vd->vdev_guid != aux_guid) { 16686f793812SPavel Zakharov boolean_t mismatch = B_FALSE; 16696f793812SPavel Zakharov if (spa->spa_trust_config && !spa->spa_extreme_rewind) { 16706f793812SPavel Zakharov if (vd != vd->vdev_top || vd->vdev_guid != top_guid) 16716f793812SPavel Zakharov mismatch = B_TRUE; 16726f793812SPavel Zakharov } else { 16736f793812SPavel Zakharov if (vd->vdev_guid != top_guid && 16746f793812SPavel Zakharov vd->vdev_top->vdev_guid != guid) 16756f793812SPavel Zakharov mismatch = B_TRUE; 1676560e6e96Seschrock } 1677560e6e96Seschrock 16786f793812SPavel Zakharov if (mismatch) { 1679560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 1680560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 1681560e6e96Seschrock nvlist_free(label); 16826f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: config guid " 16836f793812SPavel Zakharov "doesn't match label guid"); 16846f793812SPavel Zakharov vdev_dbgmsg(vd, "CONFIG: guid %llu, top_guid %llu", 16856f793812SPavel Zakharov (u_longlong_t)vd->vdev_guid, 16866f793812SPavel Zakharov (u_longlong_t)vd->vdev_top->vdev_guid); 16876f793812SPavel Zakharov vdev_dbgmsg(vd, "LABEL: guid %llu, top_guid %llu, " 16886f793812SPavel Zakharov "aux_guid %llu", (u_longlong_t)guid, 16896f793812SPavel Zakharov (u_longlong_t)top_guid, (u_longlong_t)aux_guid); 1690560e6e96Seschrock return (0); 1691560e6e96Seschrock } 16926f793812SPavel Zakharov } 1693560e6e96Seschrock 16946f793812SPavel Zakharov if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, 16956f793812SPavel Zakharov &state) != 0) { 16966f793812SPavel Zakharov vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 16976f793812SPavel Zakharov VDEV_AUX_CORRUPT_DATA); 1698560e6e96Seschrock nvlist_free(label); 16996f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: '%s' missing from label", 17006f793812SPavel Zakharov ZPOOL_CONFIG_POOL_STATE); 17016f793812SPavel Zakharov return (0); 17026f793812SPavel Zakharov } 1703560e6e96Seschrock 17046f793812SPavel Zakharov nvlist_free(label); 17056f793812SPavel Zakharov 17066f793812SPavel Zakharov /* 17076f793812SPavel Zakharov * If this is a verbatim import, no need to check the 17086f793812SPavel Zakharov * state of the pool. 17096f793812SPavel Zakharov */ 17106f793812SPavel Zakharov if (!(spa->spa_import_flags & ZFS_IMPORT_VERBATIM) && 17116f793812SPavel Zakharov spa_load_state(spa) == SPA_LOAD_OPEN && 17126f793812SPavel Zakharov state != POOL_STATE_ACTIVE) { 17136f793812SPavel Zakharov vdev_dbgmsg(vd, "vdev_validate: invalid pool state (%llu) " 17146f793812SPavel Zakharov "for spa %s", (u_longlong_t)state, spa->spa_name); 17156f793812SPavel Zakharov return (SET_ERROR(EBADF)); 17166f793812SPavel Zakharov } 17176f793812SPavel Zakharov 17186f793812SPavel Zakharov /* 17196f793812SPavel Zakharov * If we were able to open and validate a vdev that was 17206f793812SPavel Zakharov * previously marked permanently unavailable, clear that state 17216f793812SPavel Zakharov * now. 17226f793812SPavel Zakharov */ 17236f793812SPavel Zakharov if (vd->vdev_not_present) 17246f793812SPavel Zakharov vd->vdev_not_present = 0; 17256f793812SPavel Zakharov 17266f793812SPavel Zakharov return (0); 17276f793812SPavel Zakharov } 17286f793812SPavel Zakharov 17296f793812SPavel Zakharov static void 17306f793812SPavel Zakharov vdev_copy_path_impl(vdev_t *svd, vdev_t *dvd) 17316f793812SPavel Zakharov { 17326f793812SPavel Zakharov if (svd->vdev_path != NULL && dvd->vdev_path != NULL) { 17336f793812SPavel Zakharov if (strcmp(svd->vdev_path, dvd->vdev_path) != 0) { 17346f793812SPavel Zakharov zfs_dbgmsg("vdev_copy_path: vdev %llu: path changed " 17356f793812SPavel Zakharov "from '%s' to '%s'", (u_longlong_t)dvd->vdev_guid, 17366f793812SPavel Zakharov dvd->vdev_path, svd->vdev_path); 17376f793812SPavel Zakharov spa_strfree(dvd->vdev_path); 17386f793812SPavel Zakharov dvd->vdev_path = spa_strdup(svd->vdev_path); 17393ee8c80cSPavel Zakharov } 17406f793812SPavel Zakharov } else if (svd->vdev_path != NULL) { 17416f793812SPavel Zakharov dvd->vdev_path = spa_strdup(svd->vdev_path); 17426f793812SPavel Zakharov zfs_dbgmsg("vdev_copy_path: vdev %llu: path set to '%s'", 17436f793812SPavel Zakharov (u_longlong_t)dvd->vdev_guid, dvd->vdev_path); 17446f793812SPavel Zakharov } 17456f793812SPavel Zakharov } 1746560e6e96Seschrock 17476f793812SPavel Zakharov /* 17486f793812SPavel Zakharov * Recursively copy vdev paths from one vdev to another. Source and destination 17496f793812SPavel Zakharov * vdev trees must have same geometry otherwise return error. Intended to copy 17506f793812SPavel Zakharov * paths from userland config into MOS config. 17516f793812SPavel Zakharov */ 17526f793812SPavel Zakharov int 17536f793812SPavel Zakharov vdev_copy_path_strict(vdev_t *svd, vdev_t *dvd) 17546f793812SPavel Zakharov { 17556f793812SPavel Zakharov if ((svd->vdev_ops == &vdev_missing_ops) || 17566f793812SPavel Zakharov (svd->vdev_ishole && dvd->vdev_ishole) || 17576f793812SPavel Zakharov (dvd->vdev_ops == &vdev_indirect_ops)) 17586f793812SPavel Zakharov return (0); 17596f793812SPavel Zakharov 17606f793812SPavel Zakharov if (svd->vdev_ops != dvd->vdev_ops) { 17616f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: vdev type mismatch: %s != %s", 17626f793812SPavel Zakharov svd->vdev_ops->vdev_op_type, dvd->vdev_ops->vdev_op_type); 17636f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 17646f793812SPavel Zakharov } 17656f793812SPavel Zakharov 17666f793812SPavel Zakharov if (svd->vdev_guid != dvd->vdev_guid) { 17676f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: guids mismatch (%llu != " 17686f793812SPavel Zakharov "%llu)", (u_longlong_t)svd->vdev_guid, 17696f793812SPavel Zakharov (u_longlong_t)dvd->vdev_guid); 17706f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 17716f793812SPavel Zakharov } 17726f793812SPavel Zakharov 17736f793812SPavel Zakharov if (svd->vdev_children != dvd->vdev_children) { 17746f793812SPavel Zakharov vdev_dbgmsg(svd, "vdev_copy_path: children count mismatch: " 17756f793812SPavel Zakharov "%llu != %llu", (u_longlong_t)svd->vdev_children, 17766f793812SPavel Zakharov (u_longlong_t)dvd->vdev_children); 17776f793812SPavel Zakharov return (SET_ERROR(EINVAL)); 177851ece835Seschrock } 1779560e6e96Seschrock 17806f793812SPavel Zakharov for (uint64_t i = 0; i < svd->vdev_children; i++) { 17816f793812SPavel Zakharov int error = vdev_copy_path_strict(svd->vdev_child[i], 17826f793812SPavel Zakharov dvd->vdev_child[i]); 17836f793812SPavel Zakharov if (error != 0) 17846f793812SPavel Zakharov return (error); 17856f793812SPavel Zakharov } 17866f793812SPavel Zakharov 17876f793812SPavel Zakharov if (svd->vdev_ops->vdev_op_leaf) 17886f793812SPavel Zakharov vdev_copy_path_impl(svd, dvd); 17896f793812SPavel Zakharov 1790560e6e96Seschrock return (0); 1791560e6e96Seschrock } 1792560e6e96Seschrock 17936f793812SPavel Zakharov static void 17946f793812SPavel Zakharov vdev_copy_path_search(vdev_t *stvd, vdev_t *dvd) 17956f793812SPavel Zakharov { 17966f793812SPavel Zakharov ASSERT(stvd->vdev_top == stvd); 17976f793812SPavel Zakharov ASSERT3U(stvd->vdev_id, ==, dvd->vdev_top->vdev_id); 17986f793812SPavel Zakharov 17996f793812SPavel Zakharov for (uint64_t i = 0; i < dvd->vdev_children; i++) { 18006f793812SPavel Zakharov vdev_copy_path_search(stvd, dvd->vdev_child[i]); 18016f793812SPavel Zakharov } 18026f793812SPavel Zakharov 18036f793812SPavel Zakharov if (!dvd->vdev_ops->vdev_op_leaf || !vdev_is_concrete(dvd)) 18046f793812SPavel Zakharov return; 18056f793812SPavel Zakharov 18066f793812SPavel Zakharov /* 18076f793812SPavel Zakharov * The idea here is that while a vdev can shift positions within 18086f793812SPavel Zakharov * a top vdev (when replacing, attaching mirror, etc.) it cannot 18096f793812SPavel Zakharov * step outside of it. 18106f793812SPavel Zakharov */ 18116f793812SPavel Zakharov vdev_t *vd = vdev_lookup_by_guid(stvd, dvd->vdev_guid); 18126f793812SPavel Zakharov 18136f793812SPavel Zakharov if (vd == NULL || vd->vdev_ops != dvd->vdev_ops) 18146f793812SPavel Zakharov return; 18156f793812SPavel Zakharov 18166f793812SPavel Zakharov ASSERT(vd->vdev_ops->vdev_op_leaf); 18176f793812SPavel Zakharov 18186f793812SPavel Zakharov vdev_copy_path_impl(vd, dvd); 18196f793812SPavel Zakharov } 18206f793812SPavel Zakharov 18216f793812SPavel Zakharov /* 18226f793812SPavel Zakharov * Recursively copy vdev paths from one root vdev to another. Source and 18236f793812SPavel Zakharov * destination vdev trees may differ in geometry. For each destination leaf 18246f793812SPavel Zakharov * vdev, search a vdev with the same guid and top vdev id in the source. 18256f793812SPavel Zakharov * Intended to copy paths from userland config into MOS config. 18266f793812SPavel Zakharov */ 18276f793812SPavel Zakharov void 18286f793812SPavel Zakharov vdev_copy_path_relaxed(vdev_t *srvd, vdev_t *drvd) 18296f793812SPavel Zakharov { 18306f793812SPavel Zakharov uint64_t children = MIN(srvd->vdev_children, drvd->vdev_children); 18316f793812SPavel Zakharov ASSERT(srvd->vdev_ops == &vdev_root_ops); 18326f793812SPavel Zakharov ASSERT(drvd->vdev_ops == &vdev_root_ops); 18336f793812SPavel Zakharov 18346f793812SPavel Zakharov for (uint64_t i = 0; i < children; i++) { 18356f793812SPavel Zakharov vdev_copy_path_search(srvd->vdev_child[i], 18366f793812SPavel Zakharov drvd->vdev_child[i]); 18376f793812SPavel Zakharov } 18386f793812SPavel Zakharov } 18396f793812SPavel Zakharov 1840fa9e4066Sahrens /* 1841fa9e4066Sahrens * Close a virtual device. 1842fa9e4066Sahrens */ 1843fa9e4066Sahrens void 1844fa9e4066Sahrens vdev_close(vdev_t *vd) 1845fa9e4066Sahrens { 18468ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 1847095bcd66SGeorge Wilson vdev_t *pvd = vd->vdev_parent; 18488ad4d6ddSJeff Bonwick 18498ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 18508ad4d6ddSJeff Bonwick 18511195e687SMark J Musante /* 18521195e687SMark J Musante * If our parent is reopening, then we are as well, unless we are 18531195e687SMark J Musante * going offline. 18541195e687SMark J Musante */ 1855095bcd66SGeorge Wilson if (pvd != NULL && pvd->vdev_reopening) 18561195e687SMark J Musante vd->vdev_reopening = (pvd->vdev_reopening && !vd->vdev_offline); 1857095bcd66SGeorge Wilson 1858fa9e4066Sahrens vd->vdev_ops->vdev_op_close(vd); 1859fa9e4066Sahrens 18603d7072f8Seschrock vdev_cache_purge(vd); 1861fa9e4066Sahrens 1862560e6e96Seschrock /* 1863573ca77eSGeorge Wilson * We record the previous state before we close it, so that if we are 1864560e6e96Seschrock * doing a reopen(), we don't generate FMA ereports if we notice that 1865560e6e96Seschrock * it's still faulted. 1866560e6e96Seschrock */ 1867560e6e96Seschrock vd->vdev_prevstate = vd->vdev_state; 1868560e6e96Seschrock 1869fa9e4066Sahrens if (vd->vdev_offline) 1870fa9e4066Sahrens vd->vdev_state = VDEV_STATE_OFFLINE; 1871fa9e4066Sahrens else 1872fa9e4066Sahrens vd->vdev_state = VDEV_STATE_CLOSED; 1873ea8dc4b6Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 1874fa9e4066Sahrens } 1875fa9e4066Sahrens 1876dcba9f3fSGeorge Wilson void 1877dcba9f3fSGeorge Wilson vdev_hold(vdev_t *vd) 1878dcba9f3fSGeorge Wilson { 1879dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1880dcba9f3fSGeorge Wilson 1881dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1882dcba9f3fSGeorge Wilson if (spa->spa_state == POOL_STATE_UNINITIALIZED) 1883dcba9f3fSGeorge Wilson return; 1884dcba9f3fSGeorge Wilson 1885dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1886dcba9f3fSGeorge Wilson vdev_hold(vd->vdev_child[c]); 1887dcba9f3fSGeorge Wilson 1888dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1889dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_hold(vd); 1890dcba9f3fSGeorge Wilson } 1891dcba9f3fSGeorge Wilson 1892dcba9f3fSGeorge Wilson void 1893dcba9f3fSGeorge Wilson vdev_rele(vdev_t *vd) 1894dcba9f3fSGeorge Wilson { 1895dcba9f3fSGeorge Wilson spa_t *spa = vd->vdev_spa; 1896dcba9f3fSGeorge Wilson 1897dcba9f3fSGeorge Wilson ASSERT(spa_is_root(spa)); 1898dcba9f3fSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 1899dcba9f3fSGeorge Wilson vdev_rele(vd->vdev_child[c]); 1900dcba9f3fSGeorge Wilson 1901dcba9f3fSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 1902dcba9f3fSGeorge Wilson vd->vdev_ops->vdev_op_rele(vd); 1903dcba9f3fSGeorge Wilson } 1904dcba9f3fSGeorge Wilson 1905095bcd66SGeorge Wilson /* 1906095bcd66SGeorge Wilson * Reopen all interior vdevs and any unopened leaves. We don't actually 1907095bcd66SGeorge Wilson * reopen leaf vdevs which had previously been opened as they might deadlock 1908095bcd66SGeorge Wilson * on the spa_config_lock. Instead we only obtain the leaf's physical size. 1909095bcd66SGeorge Wilson * If the leaf has never been opened then open it, as usual. 1910095bcd66SGeorge Wilson */ 1911fa9e4066Sahrens void 1912ea8dc4b6Seschrock vdev_reopen(vdev_t *vd) 1913fa9e4066Sahrens { 1914ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 1915fa9e4066Sahrens 1916e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 1917ea8dc4b6Seschrock 19181195e687SMark J Musante /* set the reopening flag unless we're taking the vdev offline */ 19191195e687SMark J Musante vd->vdev_reopening = !vd->vdev_offline; 1920fa9e4066Sahrens vdev_close(vd); 1921fa9e4066Sahrens (void) vdev_open(vd); 1922fa9e4066Sahrens 192339c23413Seschrock /* 192439c23413Seschrock * Call vdev_validate() here to make sure we have the same device. 192539c23413Seschrock * Otherwise, a device with an invalid label could be successfully 192639c23413Seschrock * opened in response to vdev_reopen(). 192739c23413Seschrock */ 1928c5904d13Seschrock if (vd->vdev_aux) { 1929c5904d13Seschrock (void) vdev_validate_aux(vd); 1930e14bb325SJeff Bonwick if (vdev_readable(vd) && vdev_writeable(vd) && 19316809eb4eSEric Schrock vd->vdev_aux == &spa->spa_l2cache && 1932573ca77eSGeorge Wilson !l2arc_vdev_present(vd)) 1933573ca77eSGeorge Wilson l2arc_add_vdev(spa, vd); 1934c5904d13Seschrock } else { 19356f793812SPavel Zakharov (void) vdev_validate(vd); 1936c5904d13Seschrock } 193739c23413Seschrock 1938fa9e4066Sahrens /* 19393d7072f8Seschrock * Reassess parent vdev's health. 1940fa9e4066Sahrens */ 19413d7072f8Seschrock vdev_propagate_state(vd); 1942fa9e4066Sahrens } 1943fa9e4066Sahrens 1944fa9e4066Sahrens int 194599653d4eSeschrock vdev_create(vdev_t *vd, uint64_t txg, boolean_t isreplacing) 1946fa9e4066Sahrens { 1947fa9e4066Sahrens int error; 1948fa9e4066Sahrens 1949fa9e4066Sahrens /* 1950fa9e4066Sahrens * Normally, partial opens (e.g. of a mirror) are allowed. 1951fa9e4066Sahrens * For a create, however, we want to fail the request if 1952fa9e4066Sahrens * there are any components we can't open. 1953fa9e4066Sahrens */ 1954fa9e4066Sahrens error = vdev_open(vd); 1955fa9e4066Sahrens 1956fa9e4066Sahrens if (error || vd->vdev_state != VDEV_STATE_HEALTHY) { 1957fa9e4066Sahrens vdev_close(vd); 1958fa9e4066Sahrens return (error ? error : ENXIO); 1959fa9e4066Sahrens } 1960fa9e4066Sahrens 1961fa9e4066Sahrens /* 19620713e232SGeorge Wilson * Recursively load DTLs and initialize all labels. 1963fa9e4066Sahrens */ 19640713e232SGeorge Wilson if ((error = vdev_dtl_load(vd)) != 0 || 19650713e232SGeorge Wilson (error = vdev_label_init(vd, txg, isreplacing ? 196639c23413Seschrock VDEV_LABEL_REPLACE : VDEV_LABEL_CREATE)) != 0) { 1967fa9e4066Sahrens vdev_close(vd); 1968fa9e4066Sahrens return (error); 1969fa9e4066Sahrens } 1970fa9e4066Sahrens 1971fa9e4066Sahrens return (0); 1972fa9e4066Sahrens } 1973fa9e4066Sahrens 19740e34b6a7Sbonwick void 1975573ca77eSGeorge Wilson vdev_metaslab_set_size(vdev_t *vd) 1976fa9e4066Sahrens { 1977*86714001SSerapheim Dimitropoulos uint64_t asize = vd->vdev_asize; 1978*86714001SSerapheim Dimitropoulos uint64_t ms_shift = 0; 1979*86714001SSerapheim Dimitropoulos 1980fa9e4066Sahrens /* 1981*86714001SSerapheim Dimitropoulos * For vdevs that are bigger than 8G the metaslab size varies in 1982*86714001SSerapheim Dimitropoulos * a way that the number of metaslabs increases in powers of two, 1983*86714001SSerapheim Dimitropoulos * linearly in terms of vdev_asize, starting from 16 metaslabs. 1984*86714001SSerapheim Dimitropoulos * So for vdev_asize of 8G we get 16 metaslabs, for 16G, we get 32, 1985*86714001SSerapheim Dimitropoulos * and so on, until we hit the maximum metaslab count limit 1986*86714001SSerapheim Dimitropoulos * [vdev_max_ms_count] from which point the metaslab count stays 1987*86714001SSerapheim Dimitropoulos * the same. 1988fa9e4066Sahrens */ 1989*86714001SSerapheim Dimitropoulos ms_shift = vdev_default_ms_shift; 1990*86714001SSerapheim Dimitropoulos 1991*86714001SSerapheim Dimitropoulos if ((asize >> ms_shift) < vdev_min_ms_count) { 1992*86714001SSerapheim Dimitropoulos /* 1993*86714001SSerapheim Dimitropoulos * For devices that are less than 8G we want to have 1994*86714001SSerapheim Dimitropoulos * exactly 16 metaslabs. We don't want less as integer 1995*86714001SSerapheim Dimitropoulos * division rounds down, so less metaslabs mean more 1996*86714001SSerapheim Dimitropoulos * wasted space. We don't want more as these vdevs are 1997*86714001SSerapheim Dimitropoulos * small and in the likely event that we are running 1998*86714001SSerapheim Dimitropoulos * out of space, the SPA will have a hard time finding 1999*86714001SSerapheim Dimitropoulos * space due to fragmentation. 2000*86714001SSerapheim Dimitropoulos */ 2001*86714001SSerapheim Dimitropoulos ms_shift = highbit64(asize / vdev_min_ms_count); 2002*86714001SSerapheim Dimitropoulos ms_shift = MAX(ms_shift, SPA_MAXBLOCKSHIFT); 2003*86714001SSerapheim Dimitropoulos 2004*86714001SSerapheim Dimitropoulos } else if ((asize >> ms_shift) > vdev_max_ms_count) { 2005*86714001SSerapheim Dimitropoulos ms_shift = highbit64(asize / vdev_max_ms_count); 2006*86714001SSerapheim Dimitropoulos } 2007*86714001SSerapheim Dimitropoulos 2008*86714001SSerapheim Dimitropoulos vd->vdev_ms_shift = ms_shift; 2009*86714001SSerapheim Dimitropoulos ASSERT3U(vd->vdev_ms_shift, >=, SPA_MAXBLOCKSHIFT); 2010fa9e4066Sahrens } 2011fa9e4066Sahrens 2012fa9e4066Sahrens void 2013ecc2d604Sbonwick vdev_dirty(vdev_t *vd, int flags, void *arg, uint64_t txg) 2014fa9e4066Sahrens { 2015ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 20165cabbc6bSPrashanth Sreenivasa /* indirect vdevs don't have metaslabs or dtls */ 20175cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd) || flags == 0); 2018ecc2d604Sbonwick ASSERT(ISP2(flags)); 2019f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 2020fa9e4066Sahrens 2021ecc2d604Sbonwick if (flags & VDD_METASLAB) 2022ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, arg, txg); 2023ecc2d604Sbonwick 2024ecc2d604Sbonwick if (flags & VDD_DTL) 2025ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_dtl_list, arg, txg); 2026ecc2d604Sbonwick 2027ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_spa->spa_vdev_txg_list, vd, txg); 2028fa9e4066Sahrens } 2029fa9e4066Sahrens 20300713e232SGeorge Wilson void 20310713e232SGeorge Wilson vdev_dirty_leaves(vdev_t *vd, int flags, uint64_t txg) 20320713e232SGeorge Wilson { 20330713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 20340713e232SGeorge Wilson vdev_dirty_leaves(vd->vdev_child[c], flags, txg); 20350713e232SGeorge Wilson 20360713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 20370713e232SGeorge Wilson vdev_dirty(vd->vdev_top, flags, vd, txg); 20380713e232SGeorge Wilson } 20390713e232SGeorge Wilson 20408ad4d6ddSJeff Bonwick /* 20418ad4d6ddSJeff Bonwick * DTLs. 20428ad4d6ddSJeff Bonwick * 20438ad4d6ddSJeff Bonwick * A vdev's DTL (dirty time log) is the set of transaction groups for which 20449fb35debSEric Taylor * the vdev has less than perfect replication. There are four kinds of DTL: 20458ad4d6ddSJeff Bonwick * 20468ad4d6ddSJeff Bonwick * DTL_MISSING: txgs for which the vdev has no valid copies of the data 20478ad4d6ddSJeff Bonwick * 20488ad4d6ddSJeff Bonwick * DTL_PARTIAL: txgs for which data is available, but not fully replicated 20498ad4d6ddSJeff Bonwick * 20508ad4d6ddSJeff Bonwick * DTL_SCRUB: the txgs that could not be repaired by the last scrub; upon 20518ad4d6ddSJeff Bonwick * scrub completion, DTL_SCRUB replaces DTL_MISSING in the range of 20528ad4d6ddSJeff Bonwick * txgs that was scrubbed. 20538ad4d6ddSJeff Bonwick * 20548ad4d6ddSJeff Bonwick * DTL_OUTAGE: txgs which cannot currently be read, whether due to 20558ad4d6ddSJeff Bonwick * persistent errors or just some device being offline. 20568ad4d6ddSJeff Bonwick * Unlike the other three, the DTL_OUTAGE map is not generally 20578ad4d6ddSJeff Bonwick * maintained; it's only computed when needed, typically to 20588ad4d6ddSJeff Bonwick * determine whether a device can be detached. 20598ad4d6ddSJeff Bonwick * 20608ad4d6ddSJeff Bonwick * For leaf vdevs, DTL_MISSING and DTL_PARTIAL are identical: the device 20618ad4d6ddSJeff Bonwick * either has the data or it doesn't. 20628ad4d6ddSJeff Bonwick * 20638ad4d6ddSJeff Bonwick * For interior vdevs such as mirror and RAID-Z the picture is more complex. 20648ad4d6ddSJeff Bonwick * A vdev's DTL_PARTIAL is the union of its children's DTL_PARTIALs, because 20658ad4d6ddSJeff Bonwick * if any child is less than fully replicated, then so is its parent. 20668ad4d6ddSJeff Bonwick * A vdev's DTL_MISSING is a modified union of its children's DTL_MISSINGs, 20678ad4d6ddSJeff Bonwick * comprising only those txgs which appear in 'maxfaults' or more children; 20688ad4d6ddSJeff Bonwick * those are the txgs we don't have enough replication to read. For example, 20698ad4d6ddSJeff Bonwick * double-parity RAID-Z can tolerate up to two missing devices (maxfaults == 2); 20708ad4d6ddSJeff Bonwick * thus, its DTL_MISSING consists of the set of txgs that appear in more than 20718ad4d6ddSJeff Bonwick * two child DTL_MISSING maps. 20728ad4d6ddSJeff Bonwick * 20738ad4d6ddSJeff Bonwick * It should be clear from the above that to compute the DTLs and outage maps 20748ad4d6ddSJeff Bonwick * for all vdevs, it suffices to know just the leaf vdevs' DTL_MISSING maps. 20758ad4d6ddSJeff Bonwick * Therefore, that is all we keep on disk. When loading the pool, or after 20768ad4d6ddSJeff Bonwick * a configuration change, we generate all other DTLs from first principles. 20778ad4d6ddSJeff Bonwick */ 2078fa9e4066Sahrens void 20798ad4d6ddSJeff Bonwick vdev_dtl_dirty(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2080fa9e4066Sahrens { 20810713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 20828ad4d6ddSJeff Bonwick 20838ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 20848ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2085f9af39baSGeorge Wilson ASSERT(spa_writeable(vd->vdev_spa)); 20868ad4d6ddSJeff Bonwick 20875cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 20880713e232SGeorge Wilson if (!range_tree_contains(rt, txg, size)) 20890713e232SGeorge Wilson range_tree_add(rt, txg, size); 20905cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 2091fa9e4066Sahrens } 2092fa9e4066Sahrens 20938ad4d6ddSJeff Bonwick boolean_t 20948ad4d6ddSJeff Bonwick vdev_dtl_contains(vdev_t *vd, vdev_dtl_type_t t, uint64_t txg, uint64_t size) 2095fa9e4066Sahrens { 20960713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 20978ad4d6ddSJeff Bonwick boolean_t dirty = B_FALSE; 2098fa9e4066Sahrens 20998ad4d6ddSJeff Bonwick ASSERT(t < DTL_TYPES); 21008ad4d6ddSJeff Bonwick ASSERT(vd != vd->vdev_spa->spa_root_vdev); 2101fa9e4066Sahrens 21025cabbc6bSPrashanth Sreenivasa /* 21035cabbc6bSPrashanth Sreenivasa * While we are loading the pool, the DTLs have not been loaded yet. 21045cabbc6bSPrashanth Sreenivasa * Ignore the DTLs and try all devices. This avoids a recursive 21055cabbc6bSPrashanth Sreenivasa * mutex enter on the vdev_dtl_lock, and also makes us try hard 21065cabbc6bSPrashanth Sreenivasa * when loading the pool (relying on the checksum to ensure that 21075cabbc6bSPrashanth Sreenivasa * we get the right data -- note that we while loading, we are 21085cabbc6bSPrashanth Sreenivasa * only reading the MOS, which is always checksummed). 21095cabbc6bSPrashanth Sreenivasa */ 21105cabbc6bSPrashanth Sreenivasa if (vd->vdev_spa->spa_load_state != SPA_LOAD_NONE) 21115cabbc6bSPrashanth Sreenivasa return (B_FALSE); 21125cabbc6bSPrashanth Sreenivasa 21135cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 2114*86714001SSerapheim Dimitropoulos if (!range_tree_is_empty(rt)) 21150713e232SGeorge Wilson dirty = range_tree_contains(rt, txg, size); 21165cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 2117fa9e4066Sahrens 2118fa9e4066Sahrens return (dirty); 2119fa9e4066Sahrens } 2120fa9e4066Sahrens 21218ad4d6ddSJeff Bonwick boolean_t 21228ad4d6ddSJeff Bonwick vdev_dtl_empty(vdev_t *vd, vdev_dtl_type_t t) 21238ad4d6ddSJeff Bonwick { 21240713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[t]; 21258ad4d6ddSJeff Bonwick boolean_t empty; 21268ad4d6ddSJeff Bonwick 21275cabbc6bSPrashanth Sreenivasa mutex_enter(&vd->vdev_dtl_lock); 2128*86714001SSerapheim Dimitropoulos empty = range_tree_is_empty(rt); 21295cabbc6bSPrashanth Sreenivasa mutex_exit(&vd->vdev_dtl_lock); 21308ad4d6ddSJeff Bonwick 21318ad4d6ddSJeff Bonwick return (empty); 21328ad4d6ddSJeff Bonwick } 21338ad4d6ddSJeff Bonwick 2134b4952e17SGeorge Wilson /* 2135b4952e17SGeorge Wilson * Returns the lowest txg in the DTL range. 2136b4952e17SGeorge Wilson */ 2137b4952e17SGeorge Wilson static uint64_t 2138b4952e17SGeorge Wilson vdev_dtl_min(vdev_t *vd) 2139b4952e17SGeorge Wilson { 21400713e232SGeorge Wilson range_seg_t *rs; 2141b4952e17SGeorge Wilson 2142b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 21430713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2144b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2145b4952e17SGeorge Wilson 21460713e232SGeorge Wilson rs = avl_first(&vd->vdev_dtl[DTL_MISSING]->rt_root); 21470713e232SGeorge Wilson return (rs->rs_start - 1); 2148b4952e17SGeorge Wilson } 2149b4952e17SGeorge Wilson 2150b4952e17SGeorge Wilson /* 2151b4952e17SGeorge Wilson * Returns the highest txg in the DTL. 2152b4952e17SGeorge Wilson */ 2153b4952e17SGeorge Wilson static uint64_t 2154b4952e17SGeorge Wilson vdev_dtl_max(vdev_t *vd) 2155b4952e17SGeorge Wilson { 21560713e232SGeorge Wilson range_seg_t *rs; 2157b4952e17SGeorge Wilson 2158b4952e17SGeorge Wilson ASSERT(MUTEX_HELD(&vd->vdev_dtl_lock)); 21590713e232SGeorge Wilson ASSERT3U(range_tree_space(vd->vdev_dtl[DTL_MISSING]), !=, 0); 2160b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2161b4952e17SGeorge Wilson 21620713e232SGeorge Wilson rs = avl_last(&vd->vdev_dtl[DTL_MISSING]->rt_root); 21630713e232SGeorge Wilson return (rs->rs_end); 2164b4952e17SGeorge Wilson } 2165b4952e17SGeorge Wilson 2166b4952e17SGeorge Wilson /* 2167b4952e17SGeorge Wilson * Determine if a resilvering vdev should remove any DTL entries from 2168b4952e17SGeorge Wilson * its range. If the vdev was resilvering for the entire duration of the 2169b4952e17SGeorge Wilson * scan then it should excise that range from its DTLs. Otherwise, this 2170b4952e17SGeorge Wilson * vdev is considered partially resilvered and should leave its DTL 2171b4952e17SGeorge Wilson * entries intact. The comment in vdev_dtl_reassess() describes how we 2172b4952e17SGeorge Wilson * excise the DTLs. 2173b4952e17SGeorge Wilson */ 2174b4952e17SGeorge Wilson static boolean_t 2175b4952e17SGeorge Wilson vdev_dtl_should_excise(vdev_t *vd) 2176b4952e17SGeorge Wilson { 2177b4952e17SGeorge Wilson spa_t *spa = vd->vdev_spa; 2178b4952e17SGeorge Wilson dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 2179b4952e17SGeorge Wilson 2180b4952e17SGeorge Wilson ASSERT0(scn->scn_phys.scn_errors); 2181b4952e17SGeorge Wilson ASSERT0(vd->vdev_children); 2182b4952e17SGeorge Wilson 21832d2f193aSMatthew Ahrens if (vd->vdev_state < VDEV_STATE_DEGRADED) 21842d2f193aSMatthew Ahrens return (B_FALSE); 21852d2f193aSMatthew Ahrens 2186b4952e17SGeorge Wilson if (vd->vdev_resilver_txg == 0 || 2187*86714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_MISSING])) 2188b4952e17SGeorge Wilson return (B_TRUE); 2189b4952e17SGeorge Wilson 2190b4952e17SGeorge Wilson /* 2191b4952e17SGeorge Wilson * When a resilver is initiated the scan will assign the scn_max_txg 2192b4952e17SGeorge Wilson * value to the highest txg value that exists in all DTLs. If this 2193b4952e17SGeorge Wilson * device's max DTL is not part of this scan (i.e. it is not in 2194b4952e17SGeorge Wilson * the range (scn_min_txg, scn_max_txg] then it is not eligible 2195b4952e17SGeorge Wilson * for excision. 2196b4952e17SGeorge Wilson */ 2197b4952e17SGeorge Wilson if (vdev_dtl_max(vd) <= scn->scn_phys.scn_max_txg) { 2198b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <=, vdev_dtl_min(vd)); 2199b4952e17SGeorge Wilson ASSERT3U(scn->scn_phys.scn_min_txg, <, vd->vdev_resilver_txg); 2200b4952e17SGeorge Wilson ASSERT3U(vd->vdev_resilver_txg, <=, scn->scn_phys.scn_max_txg); 2201b4952e17SGeorge Wilson return (B_TRUE); 2202b4952e17SGeorge Wilson } 2203b4952e17SGeorge Wilson return (B_FALSE); 2204b4952e17SGeorge Wilson } 2205b4952e17SGeorge Wilson 2206fa9e4066Sahrens /* 2207fa9e4066Sahrens * Reassess DTLs after a config change or scrub completion. 2208fa9e4066Sahrens */ 2209fa9e4066Sahrens void 2210fa9e4066Sahrens vdev_dtl_reassess(vdev_t *vd, uint64_t txg, uint64_t scrub_txg, int scrub_done) 2211fa9e4066Sahrens { 2212ea8dc4b6Seschrock spa_t *spa = vd->vdev_spa; 22138ad4d6ddSJeff Bonwick avl_tree_t reftree; 22148ad4d6ddSJeff Bonwick int minref; 2215fa9e4066Sahrens 22168ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 2217fa9e4066Sahrens 22188ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 22198ad4d6ddSJeff Bonwick vdev_dtl_reassess(vd->vdev_child[c], txg, 22208ad4d6ddSJeff Bonwick scrub_txg, scrub_done); 22218ad4d6ddSJeff Bonwick 22225cabbc6bSPrashanth Sreenivasa if (vd == spa->spa_root_vdev || !vdev_is_concrete(vd) || vd->vdev_aux) 22238ad4d6ddSJeff Bonwick return; 22248ad4d6ddSJeff Bonwick 22258ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 22263f9d6ad7SLin Ling dsl_scan_t *scn = spa->spa_dsl_pool->dp_scan; 22273f9d6ad7SLin Ling 2228fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 2229b4952e17SGeorge Wilson 2230b4952e17SGeorge Wilson /* 2231b4952e17SGeorge Wilson * If we've completed a scan cleanly then determine 2232b4952e17SGeorge Wilson * if this vdev should remove any DTLs. We only want to 2233b4952e17SGeorge Wilson * excise regions on vdevs that were available during 2234b4952e17SGeorge Wilson * the entire duration of this scan. 2235b4952e17SGeorge Wilson */ 2236088f3894Sahrens if (scrub_txg != 0 && 22373f9d6ad7SLin Ling (spa->spa_scrub_started || 2238b4952e17SGeorge Wilson (scn != NULL && scn->scn_phys.scn_errors == 0)) && 2239b4952e17SGeorge Wilson vdev_dtl_should_excise(vd)) { 2240088f3894Sahrens /* 2241088f3894Sahrens * We completed a scrub up to scrub_txg. If we 2242088f3894Sahrens * did it without rebooting, then the scrub dtl 2243088f3894Sahrens * will be valid, so excise the old region and 2244088f3894Sahrens * fold in the scrub dtl. Otherwise, leave the 2245088f3894Sahrens * dtl as-is if there was an error. 22468ad4d6ddSJeff Bonwick * 22478ad4d6ddSJeff Bonwick * There's little trick here: to excise the beginning 22488ad4d6ddSJeff Bonwick * of the DTL_MISSING map, we put it into a reference 22498ad4d6ddSJeff Bonwick * tree and then add a segment with refcnt -1 that 22508ad4d6ddSJeff Bonwick * covers the range [0, scrub_txg). This means 22518ad4d6ddSJeff Bonwick * that each txg in that range has refcnt -1 or 0. 22528ad4d6ddSJeff Bonwick * We then add DTL_SCRUB with a refcnt of 2, so that 22538ad4d6ddSJeff Bonwick * entries in the range [0, scrub_txg) will have a 22548ad4d6ddSJeff Bonwick * positive refcnt -- either 1 or 2. We then convert 22558ad4d6ddSJeff Bonwick * the reference tree into the new DTL_MISSING map. 2256088f3894Sahrens */ 22570713e232SGeorge Wilson space_reftree_create(&reftree); 22580713e232SGeorge Wilson space_reftree_add_map(&reftree, 22590713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 22600713e232SGeorge Wilson space_reftree_add_seg(&reftree, 0, scrub_txg, -1); 22610713e232SGeorge Wilson space_reftree_add_map(&reftree, 22620713e232SGeorge Wilson vd->vdev_dtl[DTL_SCRUB], 2); 22630713e232SGeorge Wilson space_reftree_generate_map(&reftree, 22640713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], 1); 22650713e232SGeorge Wilson space_reftree_destroy(&reftree); 2266fa9e4066Sahrens } 22670713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_PARTIAL], NULL, NULL); 22680713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 22690713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_PARTIAL]); 2270fa9e4066Sahrens if (scrub_done) 22710713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_SCRUB], NULL, NULL); 22720713e232SGeorge Wilson range_tree_vacate(vd->vdev_dtl[DTL_OUTAGE], NULL, NULL); 22738ad4d6ddSJeff Bonwick if (!vdev_readable(vd)) 22740713e232SGeorge Wilson range_tree_add(vd->vdev_dtl[DTL_OUTAGE], 0, -1ULL); 22758ad4d6ddSJeff Bonwick else 22760713e232SGeorge Wilson range_tree_walk(vd->vdev_dtl[DTL_MISSING], 22770713e232SGeorge Wilson range_tree_add, vd->vdev_dtl[DTL_OUTAGE]); 2278b4952e17SGeorge Wilson 2279b4952e17SGeorge Wilson /* 2280b4952e17SGeorge Wilson * If the vdev was resilvering and no longer has any 2281b4952e17SGeorge Wilson * DTLs then reset its resilvering flag. 2282b4952e17SGeorge Wilson */ 2283b4952e17SGeorge Wilson if (vd->vdev_resilver_txg != 0 && 2284*86714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 2285*86714001SSerapheim Dimitropoulos range_tree_is_empty(vd->vdev_dtl[DTL_OUTAGE])) 2286b4952e17SGeorge Wilson vd->vdev_resilver_txg = 0; 2287b4952e17SGeorge Wilson 2288fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2289088f3894Sahrens 2290ecc2d604Sbonwick if (txg != 0) 2291ecc2d604Sbonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, txg); 2292fa9e4066Sahrens return; 2293fa9e4066Sahrens } 2294fa9e4066Sahrens 2295fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 22968ad4d6ddSJeff Bonwick for (int t = 0; t < DTL_TYPES; t++) { 229799bb17e2SEric Taylor /* account for child's outage in parent's missing map */ 229899bb17e2SEric Taylor int s = (t == DTL_MISSING) ? DTL_OUTAGE: t; 22998ad4d6ddSJeff Bonwick if (t == DTL_SCRUB) 23008ad4d6ddSJeff Bonwick continue; /* leaf vdevs only */ 23018ad4d6ddSJeff Bonwick if (t == DTL_PARTIAL) 23028ad4d6ddSJeff Bonwick minref = 1; /* i.e. non-zero */ 23038ad4d6ddSJeff Bonwick else if (vd->vdev_nparity != 0) 23048ad4d6ddSJeff Bonwick minref = vd->vdev_nparity + 1; /* RAID-Z */ 23058ad4d6ddSJeff Bonwick else 23068ad4d6ddSJeff Bonwick minref = vd->vdev_children; /* any kind of mirror */ 23070713e232SGeorge Wilson space_reftree_create(&reftree); 23088ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 23098ad4d6ddSJeff Bonwick vdev_t *cvd = vd->vdev_child[c]; 23108ad4d6ddSJeff Bonwick mutex_enter(&cvd->vdev_dtl_lock); 23110713e232SGeorge Wilson space_reftree_add_map(&reftree, cvd->vdev_dtl[s], 1); 23128ad4d6ddSJeff Bonwick mutex_exit(&cvd->vdev_dtl_lock); 23138ad4d6ddSJeff Bonwick } 23140713e232SGeorge Wilson space_reftree_generate_map(&reftree, vd->vdev_dtl[t], minref); 23150713e232SGeorge Wilson space_reftree_destroy(&reftree); 2316fa9e4066Sahrens } 23178ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_dtl_lock); 2318fa9e4066Sahrens } 2319fa9e4066Sahrens 23200713e232SGeorge Wilson int 2321fa9e4066Sahrens vdev_dtl_load(vdev_t *vd) 2322fa9e4066Sahrens { 2323fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2324ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 23250713e232SGeorge Wilson int error = 0; 2326fa9e4066Sahrens 23270713e232SGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && vd->vdev_dtl_object != 0) { 23285cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 2329fa9e4066Sahrens 23300713e232SGeorge Wilson error = space_map_open(&vd->vdev_dtl_sm, mos, 23315cabbc6bSPrashanth Sreenivasa vd->vdev_dtl_object, 0, -1ULL, 0); 23320713e232SGeorge Wilson if (error) 23330713e232SGeorge Wilson return (error); 23340713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2335fa9e4066Sahrens 23360713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 233788ecc943SGeorge Wilson 23380713e232SGeorge Wilson /* 23390713e232SGeorge Wilson * Now that we've opened the space_map we need to update 23400713e232SGeorge Wilson * the in-core DTL. 23410713e232SGeorge Wilson */ 23420713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 2343ecc2d604Sbonwick 23440713e232SGeorge Wilson error = space_map_load(vd->vdev_dtl_sm, 23450713e232SGeorge Wilson vd->vdev_dtl[DTL_MISSING], SM_ALLOC); 23460713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2347fa9e4066Sahrens 23480713e232SGeorge Wilson return (error); 23490713e232SGeorge Wilson } 23500713e232SGeorge Wilson 23510713e232SGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 23520713e232SGeorge Wilson error = vdev_dtl_load(vd->vdev_child[c]); 23530713e232SGeorge Wilson if (error != 0) 23540713e232SGeorge Wilson break; 23550713e232SGeorge Wilson } 2356fa9e4066Sahrens 2357fa9e4066Sahrens return (error); 2358fa9e4066Sahrens } 2359fa9e4066Sahrens 2360215198a6SJoe Stein void 2361215198a6SJoe Stein vdev_destroy_unlink_zap(vdev_t *vd, uint64_t zapobj, dmu_tx_t *tx) 2362215198a6SJoe Stein { 2363215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2364215198a6SJoe Stein 2365215198a6SJoe Stein VERIFY0(zap_destroy(spa->spa_meta_objset, zapobj, tx)); 2366215198a6SJoe Stein VERIFY0(zap_remove_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2367215198a6SJoe Stein zapobj, tx)); 2368215198a6SJoe Stein } 2369215198a6SJoe Stein 2370215198a6SJoe Stein uint64_t 2371215198a6SJoe Stein vdev_create_link_zap(vdev_t *vd, dmu_tx_t *tx) 2372215198a6SJoe Stein { 2373215198a6SJoe Stein spa_t *spa = vd->vdev_spa; 2374215198a6SJoe Stein uint64_t zap = zap_create(spa->spa_meta_objset, DMU_OTN_ZAP_METADATA, 2375215198a6SJoe Stein DMU_OT_NONE, 0, tx); 2376215198a6SJoe Stein 2377215198a6SJoe Stein ASSERT(zap != 0); 2378215198a6SJoe Stein VERIFY0(zap_add_int(spa->spa_meta_objset, spa->spa_all_vdev_zaps, 2379215198a6SJoe Stein zap, tx)); 2380215198a6SJoe Stein 2381215198a6SJoe Stein return (zap); 2382215198a6SJoe Stein } 2383215198a6SJoe Stein 2384215198a6SJoe Stein void 2385215198a6SJoe Stein vdev_construct_zaps(vdev_t *vd, dmu_tx_t *tx) 2386215198a6SJoe Stein { 2387215198a6SJoe Stein if (vd->vdev_ops != &vdev_hole_ops && 2388215198a6SJoe Stein vd->vdev_ops != &vdev_missing_ops && 2389215198a6SJoe Stein vd->vdev_ops != &vdev_root_ops && 2390215198a6SJoe Stein !vd->vdev_top->vdev_removing) { 2391215198a6SJoe Stein if (vd->vdev_ops->vdev_op_leaf && vd->vdev_leaf_zap == 0) { 2392215198a6SJoe Stein vd->vdev_leaf_zap = vdev_create_link_zap(vd, tx); 2393215198a6SJoe Stein } 2394215198a6SJoe Stein if (vd == vd->vdev_top && vd->vdev_top_zap == 0) { 2395215198a6SJoe Stein vd->vdev_top_zap = vdev_create_link_zap(vd, tx); 2396215198a6SJoe Stein } 2397215198a6SJoe Stein } 2398215198a6SJoe Stein for (uint64_t i = 0; i < vd->vdev_children; i++) { 2399215198a6SJoe Stein vdev_construct_zaps(vd->vdev_child[i], tx); 2400215198a6SJoe Stein } 2401215198a6SJoe Stein } 2402215198a6SJoe Stein 2403fa9e4066Sahrens void 2404fa9e4066Sahrens vdev_dtl_sync(vdev_t *vd, uint64_t txg) 2405fa9e4066Sahrens { 2406fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 24070713e232SGeorge Wilson range_tree_t *rt = vd->vdev_dtl[DTL_MISSING]; 2408ecc2d604Sbonwick objset_t *mos = spa->spa_meta_objset; 24090713e232SGeorge Wilson range_tree_t *rtsync; 2410fa9e4066Sahrens dmu_tx_t *tx; 24110713e232SGeorge Wilson uint64_t object = space_map_object(vd->vdev_dtl_sm); 2412fa9e4066Sahrens 24135cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 24140713e232SGeorge Wilson ASSERT(vd->vdev_ops->vdev_op_leaf); 241588ecc943SGeorge Wilson 2416fa9e4066Sahrens tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2417fa9e4066Sahrens 24180713e232SGeorge Wilson if (vd->vdev_detached || vd->vdev_top->vdev_removing) { 24190713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 24200713e232SGeorge Wilson space_map_free(vd->vdev_dtl_sm, tx); 24210713e232SGeorge Wilson space_map_close(vd->vdev_dtl_sm); 24220713e232SGeorge Wilson vd->vdev_dtl_sm = NULL; 24230713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2424215198a6SJoe Stein 2425215198a6SJoe Stein /* 2426215198a6SJoe Stein * We only destroy the leaf ZAP for detached leaves or for 2427215198a6SJoe Stein * removed log devices. Removed data devices handle leaf ZAP 2428215198a6SJoe Stein * cleanup later, once cancellation is no longer possible. 2429215198a6SJoe Stein */ 2430215198a6SJoe Stein if (vd->vdev_leaf_zap != 0 && (vd->vdev_detached || 2431215198a6SJoe Stein vd->vdev_top->vdev_islog)) { 2432215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_leaf_zap, tx); 2433215198a6SJoe Stein vd->vdev_leaf_zap = 0; 2434215198a6SJoe Stein } 2435215198a6SJoe Stein 2436fa9e4066Sahrens dmu_tx_commit(tx); 2437fa9e4066Sahrens return; 2438fa9e4066Sahrens } 2439fa9e4066Sahrens 24400713e232SGeorge Wilson if (vd->vdev_dtl_sm == NULL) { 24410713e232SGeorge Wilson uint64_t new_object; 24420713e232SGeorge Wilson 2443*86714001SSerapheim Dimitropoulos new_object = space_map_alloc(mos, vdev_dtl_sm_blksz, tx); 24440713e232SGeorge Wilson VERIFY3U(new_object, !=, 0); 24450713e232SGeorge Wilson 24460713e232SGeorge Wilson VERIFY0(space_map_open(&vd->vdev_dtl_sm, mos, new_object, 24475cabbc6bSPrashanth Sreenivasa 0, -1ULL, 0)); 24480713e232SGeorge Wilson ASSERT(vd->vdev_dtl_sm != NULL); 2449fa9e4066Sahrens } 2450fa9e4066Sahrens 24515cabbc6bSPrashanth Sreenivasa rtsync = range_tree_create(NULL, NULL); 2452fa9e4066Sahrens 2453fa9e4066Sahrens mutex_enter(&vd->vdev_dtl_lock); 24540713e232SGeorge Wilson range_tree_walk(rt, range_tree_add, rtsync); 2455fa9e4066Sahrens mutex_exit(&vd->vdev_dtl_lock); 2456fa9e4066Sahrens 2457*86714001SSerapheim Dimitropoulos space_map_truncate(vd->vdev_dtl_sm, vdev_dtl_sm_blksz, tx); 24580713e232SGeorge Wilson space_map_write(vd->vdev_dtl_sm, rtsync, SM_ALLOC, tx); 24590713e232SGeorge Wilson range_tree_vacate(rtsync, NULL, NULL); 2460fa9e4066Sahrens 24610713e232SGeorge Wilson range_tree_destroy(rtsync); 2462fa9e4066Sahrens 24630713e232SGeorge Wilson /* 24640713e232SGeorge Wilson * If the object for the space map has changed then dirty 24650713e232SGeorge Wilson * the top level so that we update the config. 24660713e232SGeorge Wilson */ 24670713e232SGeorge Wilson if (object != space_map_object(vd->vdev_dtl_sm)) { 24683ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "txg %llu, spa %s, DTL old object %llu, " 24693ee8c80cSPavel Zakharov "new object %llu", (u_longlong_t)txg, spa_name(spa), 24703ee8c80cSPavel Zakharov (u_longlong_t)object, 24713ee8c80cSPavel Zakharov (u_longlong_t)space_map_object(vd->vdev_dtl_sm)); 24720713e232SGeorge Wilson vdev_config_dirty(vd->vdev_top); 24730713e232SGeorge Wilson } 2474fa9e4066Sahrens 2475fa9e4066Sahrens dmu_tx_commit(tx); 24760713e232SGeorge Wilson 24770713e232SGeorge Wilson mutex_enter(&vd->vdev_dtl_lock); 24780713e232SGeorge Wilson space_map_update(vd->vdev_dtl_sm); 24790713e232SGeorge Wilson mutex_exit(&vd->vdev_dtl_lock); 2480fa9e4066Sahrens } 2481fa9e4066Sahrens 24828ad4d6ddSJeff Bonwick /* 24838ad4d6ddSJeff Bonwick * Determine whether the specified vdev can be offlined/detached/removed 24848ad4d6ddSJeff Bonwick * without losing data. 24858ad4d6ddSJeff Bonwick */ 24868ad4d6ddSJeff Bonwick boolean_t 24878ad4d6ddSJeff Bonwick vdev_dtl_required(vdev_t *vd) 24888ad4d6ddSJeff Bonwick { 24898ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 24908ad4d6ddSJeff Bonwick vdev_t *tvd = vd->vdev_top; 24918ad4d6ddSJeff Bonwick uint8_t cant_read = vd->vdev_cant_read; 24928ad4d6ddSJeff Bonwick boolean_t required; 24938ad4d6ddSJeff Bonwick 24948ad4d6ddSJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 24958ad4d6ddSJeff Bonwick 24968ad4d6ddSJeff Bonwick if (vd == spa->spa_root_vdev || vd == tvd) 24978ad4d6ddSJeff Bonwick return (B_TRUE); 24988ad4d6ddSJeff Bonwick 24998ad4d6ddSJeff Bonwick /* 25008ad4d6ddSJeff Bonwick * Temporarily mark the device as unreadable, and then determine 25018ad4d6ddSJeff Bonwick * whether this results in any DTL outages in the top-level vdev. 25028ad4d6ddSJeff Bonwick * If not, we can safely offline/detach/remove the device. 25038ad4d6ddSJeff Bonwick */ 25048ad4d6ddSJeff Bonwick vd->vdev_cant_read = B_TRUE; 25058ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 25068ad4d6ddSJeff Bonwick required = !vdev_dtl_empty(tvd, DTL_OUTAGE); 25078ad4d6ddSJeff Bonwick vd->vdev_cant_read = cant_read; 25088ad4d6ddSJeff Bonwick vdev_dtl_reassess(tvd, 0, 0, B_FALSE); 25098ad4d6ddSJeff Bonwick 2510cb04b873SMark J Musante if (!required && zio_injection_enabled) 2511cb04b873SMark J Musante required = !!zio_handle_device_injection(vd, NULL, ECHILD); 2512cb04b873SMark J Musante 25138ad4d6ddSJeff Bonwick return (required); 25148ad4d6ddSJeff Bonwick } 25158ad4d6ddSJeff Bonwick 2516088f3894Sahrens /* 2517088f3894Sahrens * Determine if resilver is needed, and if so the txg range. 2518088f3894Sahrens */ 2519088f3894Sahrens boolean_t 2520088f3894Sahrens vdev_resilver_needed(vdev_t *vd, uint64_t *minp, uint64_t *maxp) 2521088f3894Sahrens { 2522088f3894Sahrens boolean_t needed = B_FALSE; 2523088f3894Sahrens uint64_t thismin = UINT64_MAX; 2524088f3894Sahrens uint64_t thismax = 0; 2525088f3894Sahrens 2526088f3894Sahrens if (vd->vdev_children == 0) { 2527088f3894Sahrens mutex_enter(&vd->vdev_dtl_lock); 2528*86714001SSerapheim Dimitropoulos if (!range_tree_is_empty(vd->vdev_dtl[DTL_MISSING]) && 25298ad4d6ddSJeff Bonwick vdev_writeable(vd)) { 2530088f3894Sahrens 2531b4952e17SGeorge Wilson thismin = vdev_dtl_min(vd); 2532b4952e17SGeorge Wilson thismax = vdev_dtl_max(vd); 2533088f3894Sahrens needed = B_TRUE; 2534088f3894Sahrens } 2535088f3894Sahrens mutex_exit(&vd->vdev_dtl_lock); 2536088f3894Sahrens } else { 25378ad4d6ddSJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) { 2538088f3894Sahrens vdev_t *cvd = vd->vdev_child[c]; 2539088f3894Sahrens uint64_t cmin, cmax; 2540088f3894Sahrens 2541088f3894Sahrens if (vdev_resilver_needed(cvd, &cmin, &cmax)) { 2542088f3894Sahrens thismin = MIN(thismin, cmin); 2543088f3894Sahrens thismax = MAX(thismax, cmax); 2544088f3894Sahrens needed = B_TRUE; 2545088f3894Sahrens } 2546088f3894Sahrens } 2547088f3894Sahrens } 2548088f3894Sahrens 2549088f3894Sahrens if (needed && minp) { 2550088f3894Sahrens *minp = thismin; 2551088f3894Sahrens *maxp = thismax; 2552088f3894Sahrens } 2553088f3894Sahrens return (needed); 2554088f3894Sahrens } 2555088f3894Sahrens 2556*86714001SSerapheim Dimitropoulos /* 2557*86714001SSerapheim Dimitropoulos * Gets the checkpoint space map object from the vdev's ZAP. 2558*86714001SSerapheim Dimitropoulos * Returns the spacemap object, or 0 if it wasn't in the ZAP 2559*86714001SSerapheim Dimitropoulos * or the ZAP doesn't exist yet. 2560*86714001SSerapheim Dimitropoulos */ 2561*86714001SSerapheim Dimitropoulos int 2562*86714001SSerapheim Dimitropoulos vdev_checkpoint_sm_object(vdev_t *vd) 2563*86714001SSerapheim Dimitropoulos { 2564*86714001SSerapheim Dimitropoulos ASSERT0(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER)); 2565*86714001SSerapheim Dimitropoulos if (vd->vdev_top_zap == 0) { 2566*86714001SSerapheim Dimitropoulos return (0); 2567*86714001SSerapheim Dimitropoulos } 2568*86714001SSerapheim Dimitropoulos 2569*86714001SSerapheim Dimitropoulos uint64_t sm_obj = 0; 2570*86714001SSerapheim Dimitropoulos int err = zap_lookup(spa_meta_objset(vd->vdev_spa), vd->vdev_top_zap, 2571*86714001SSerapheim Dimitropoulos VDEV_TOP_ZAP_POOL_CHECKPOINT_SM, sizeof (uint64_t), 1, &sm_obj); 2572*86714001SSerapheim Dimitropoulos 2573*86714001SSerapheim Dimitropoulos ASSERT(err == 0 || err == ENOENT); 2574*86714001SSerapheim Dimitropoulos 2575*86714001SSerapheim Dimitropoulos return (sm_obj); 2576*86714001SSerapheim Dimitropoulos } 2577*86714001SSerapheim Dimitropoulos 25785cabbc6bSPrashanth Sreenivasa int 2579ea8dc4b6Seschrock vdev_load(vdev_t *vd) 2580fa9e4066Sahrens { 25815cabbc6bSPrashanth Sreenivasa int error = 0; 2582fa9e4066Sahrens /* 2583fa9e4066Sahrens * Recursively load all children. 2584fa9e4066Sahrens */ 25855cabbc6bSPrashanth Sreenivasa for (int c = 0; c < vd->vdev_children; c++) { 25865cabbc6bSPrashanth Sreenivasa error = vdev_load(vd->vdev_child[c]); 25875cabbc6bSPrashanth Sreenivasa if (error != 0) { 25885cabbc6bSPrashanth Sreenivasa return (error); 25895cabbc6bSPrashanth Sreenivasa } 25905cabbc6bSPrashanth Sreenivasa } 25915cabbc6bSPrashanth Sreenivasa 25925cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vd); 2593fa9e4066Sahrens 2594fa9e4066Sahrens /* 25950e34b6a7Sbonwick * If this is a top-level vdev, initialize its metaslabs. 2596fa9e4066Sahrens */ 25975cabbc6bSPrashanth Sreenivasa if (vd == vd->vdev_top && vdev_is_concrete(vd)) { 25985cabbc6bSPrashanth Sreenivasa if (vd->vdev_ashift == 0 || vd->vdev_asize == 0) { 25995cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 26005cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 26013ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: invalid size. ashift=%llu, " 26023ee8c80cSPavel Zakharov "asize=%llu", (u_longlong_t)vd->vdev_ashift, 26033ee8c80cSPavel Zakharov (u_longlong_t)vd->vdev_asize); 26045cabbc6bSPrashanth Sreenivasa return (SET_ERROR(ENXIO)); 26055cabbc6bSPrashanth Sreenivasa } else if ((error = vdev_metaslab_init(vd, 0)) != 0) { 26063ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: metaslab_init failed " 26073ee8c80cSPavel Zakharov "[error=%d]", error); 26085cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 26095cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 26105cabbc6bSPrashanth Sreenivasa return (error); 26115cabbc6bSPrashanth Sreenivasa } 2612*86714001SSerapheim Dimitropoulos 2613*86714001SSerapheim Dimitropoulos uint64_t checkpoint_sm_obj = vdev_checkpoint_sm_object(vd); 2614*86714001SSerapheim Dimitropoulos if (checkpoint_sm_obj != 0) { 2615*86714001SSerapheim Dimitropoulos objset_t *mos = spa_meta_objset(vd->vdev_spa); 2616*86714001SSerapheim Dimitropoulos ASSERT(vd->vdev_asize != 0); 2617*86714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_checkpoint_sm, ==, NULL); 2618*86714001SSerapheim Dimitropoulos 2619*86714001SSerapheim Dimitropoulos if ((error = space_map_open(&vd->vdev_checkpoint_sm, 2620*86714001SSerapheim Dimitropoulos mos, checkpoint_sm_obj, 0, vd->vdev_asize, 2621*86714001SSerapheim Dimitropoulos vd->vdev_ashift))) { 2622*86714001SSerapheim Dimitropoulos vdev_dbgmsg(vd, "vdev_load: space_map_open " 2623*86714001SSerapheim Dimitropoulos "failed for checkpoint spacemap (obj %llu) " 2624*86714001SSerapheim Dimitropoulos "[error=%d]", 2625*86714001SSerapheim Dimitropoulos (u_longlong_t)checkpoint_sm_obj, error); 2626*86714001SSerapheim Dimitropoulos return (error); 2627*86714001SSerapheim Dimitropoulos } 2628*86714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_checkpoint_sm, !=, NULL); 2629*86714001SSerapheim Dimitropoulos space_map_update(vd->vdev_checkpoint_sm); 2630*86714001SSerapheim Dimitropoulos 2631*86714001SSerapheim Dimitropoulos /* 2632*86714001SSerapheim Dimitropoulos * Since the checkpoint_sm contains free entries 2633*86714001SSerapheim Dimitropoulos * exclusively we can use sm_alloc to indicate the 2634*86714001SSerapheim Dimitropoulos * culmulative checkpointed space that has been freed. 2635*86714001SSerapheim Dimitropoulos */ 2636*86714001SSerapheim Dimitropoulos vd->vdev_stat.vs_checkpoint_space = 2637*86714001SSerapheim Dimitropoulos -vd->vdev_checkpoint_sm->sm_alloc; 2638*86714001SSerapheim Dimitropoulos vd->vdev_spa->spa_checkpoint_info.sci_dspace += 2639*86714001SSerapheim Dimitropoulos vd->vdev_stat.vs_checkpoint_space; 2640*86714001SSerapheim Dimitropoulos } 26415cabbc6bSPrashanth Sreenivasa } 2642fa9e4066Sahrens 2643fa9e4066Sahrens /* 2644fa9e4066Sahrens * If this is a leaf vdev, load its DTL. 2645fa9e4066Sahrens */ 26465cabbc6bSPrashanth Sreenivasa if (vd->vdev_ops->vdev_op_leaf && (error = vdev_dtl_load(vd)) != 0) { 2647560e6e96Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 2648560e6e96Seschrock VDEV_AUX_CORRUPT_DATA); 26493ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: vdev_dtl_load failed " 26503ee8c80cSPavel Zakharov "[error=%d]", error); 26515cabbc6bSPrashanth Sreenivasa return (error); 26525cabbc6bSPrashanth Sreenivasa } 26535cabbc6bSPrashanth Sreenivasa 26545cabbc6bSPrashanth Sreenivasa uint64_t obsolete_sm_object = vdev_obsolete_sm_object(vd); 26555cabbc6bSPrashanth Sreenivasa if (obsolete_sm_object != 0) { 26565cabbc6bSPrashanth Sreenivasa objset_t *mos = vd->vdev_spa->spa_meta_objset; 26575cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_asize != 0); 2658*86714001SSerapheim Dimitropoulos ASSERT3P(vd->vdev_obsolete_sm, ==, NULL); 26595cabbc6bSPrashanth Sreenivasa 26605cabbc6bSPrashanth Sreenivasa if ((error = space_map_open(&vd->vdev_obsolete_sm, mos, 26615cabbc6bSPrashanth Sreenivasa obsolete_sm_object, 0, vd->vdev_asize, 0))) { 26625cabbc6bSPrashanth Sreenivasa vdev_set_state(vd, B_FALSE, VDEV_STATE_CANT_OPEN, 26635cabbc6bSPrashanth Sreenivasa VDEV_AUX_CORRUPT_DATA); 26643ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "vdev_load: space_map_open failed for " 26653ee8c80cSPavel Zakharov "obsolete spacemap (obj %llu) [error=%d]", 26663ee8c80cSPavel Zakharov (u_longlong_t)obsolete_sm_object, error); 26675cabbc6bSPrashanth Sreenivasa return (error); 26685cabbc6bSPrashanth Sreenivasa } 26695cabbc6bSPrashanth Sreenivasa space_map_update(vd->vdev_obsolete_sm); 26705cabbc6bSPrashanth Sreenivasa } 26715cabbc6bSPrashanth Sreenivasa 26725cabbc6bSPrashanth Sreenivasa return (0); 2673fa9e4066Sahrens } 2674fa9e4066Sahrens 267599653d4eSeschrock /* 2676fa94a07fSbrendan * The special vdev case is used for hot spares and l2cache devices. Its 2677fa94a07fSbrendan * sole purpose it to set the vdev state for the associated vdev. To do this, 2678fa94a07fSbrendan * we make sure that we can open the underlying device, then try to read the 2679fa94a07fSbrendan * label, and make sure that the label is sane and that it hasn't been 2680fa94a07fSbrendan * repurposed to another pool. 268199653d4eSeschrock */ 268299653d4eSeschrock int 2683fa94a07fSbrendan vdev_validate_aux(vdev_t *vd) 268499653d4eSeschrock { 268599653d4eSeschrock nvlist_t *label; 268699653d4eSeschrock uint64_t guid, version; 268799653d4eSeschrock uint64_t state; 268899653d4eSeschrock 2689e14bb325SJeff Bonwick if (!vdev_readable(vd)) 2690c5904d13Seschrock return (0); 2691c5904d13Seschrock 2692dfbb9432SGeorge Wilson if ((label = vdev_label_read_config(vd, -1ULL)) == NULL) { 269399653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 269499653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 269599653d4eSeschrock return (-1); 269699653d4eSeschrock } 269799653d4eSeschrock 269899653d4eSeschrock if (nvlist_lookup_uint64(label, ZPOOL_CONFIG_VERSION, &version) != 0 || 2699ad135b5dSChristopher Siden !SPA_VERSION_IS_SUPPORTED(version) || 270099653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_GUID, &guid) != 0 || 270199653d4eSeschrock guid != vd->vdev_guid || 270299653d4eSeschrock nvlist_lookup_uint64(label, ZPOOL_CONFIG_POOL_STATE, &state) != 0) { 270399653d4eSeschrock vdev_set_state(vd, B_TRUE, VDEV_STATE_CANT_OPEN, 270499653d4eSeschrock VDEV_AUX_CORRUPT_DATA); 270599653d4eSeschrock nvlist_free(label); 270699653d4eSeschrock return (-1); 270799653d4eSeschrock } 270899653d4eSeschrock 270999653d4eSeschrock /* 271099653d4eSeschrock * We don't actually check the pool state here. If it's in fact in 271199653d4eSeschrock * use by another pool, we update this fact on the fly when requested. 271299653d4eSeschrock */ 271399653d4eSeschrock nvlist_free(label); 271499653d4eSeschrock return (0); 271599653d4eSeschrock } 271699653d4eSeschrock 27175cabbc6bSPrashanth Sreenivasa /* 27185cabbc6bSPrashanth Sreenivasa * Free the objects used to store this vdev's spacemaps, and the array 27195cabbc6bSPrashanth Sreenivasa * that points to them. 27205cabbc6bSPrashanth Sreenivasa */ 272188ecc943SGeorge Wilson void 27225cabbc6bSPrashanth Sreenivasa vdev_destroy_spacemaps(vdev_t *vd, dmu_tx_t *tx) 27235cabbc6bSPrashanth Sreenivasa { 27245cabbc6bSPrashanth Sreenivasa if (vd->vdev_ms_array == 0) 27255cabbc6bSPrashanth Sreenivasa return; 27265cabbc6bSPrashanth Sreenivasa 27275cabbc6bSPrashanth Sreenivasa objset_t *mos = vd->vdev_spa->spa_meta_objset; 27285cabbc6bSPrashanth Sreenivasa uint64_t array_count = vd->vdev_asize >> vd->vdev_ms_shift; 27295cabbc6bSPrashanth Sreenivasa size_t array_bytes = array_count * sizeof (uint64_t); 27305cabbc6bSPrashanth Sreenivasa uint64_t *smobj_array = kmem_alloc(array_bytes, KM_SLEEP); 27315cabbc6bSPrashanth Sreenivasa VERIFY0(dmu_read(mos, vd->vdev_ms_array, 0, 27325cabbc6bSPrashanth Sreenivasa array_bytes, smobj_array, 0)); 27335cabbc6bSPrashanth Sreenivasa 27345cabbc6bSPrashanth Sreenivasa for (uint64_t i = 0; i < array_count; i++) { 27355cabbc6bSPrashanth Sreenivasa uint64_t smobj = smobj_array[i]; 27365cabbc6bSPrashanth Sreenivasa if (smobj == 0) 27375cabbc6bSPrashanth Sreenivasa continue; 27385cabbc6bSPrashanth Sreenivasa 27395cabbc6bSPrashanth Sreenivasa space_map_free_obj(mos, smobj, tx); 27405cabbc6bSPrashanth Sreenivasa } 27415cabbc6bSPrashanth Sreenivasa 27425cabbc6bSPrashanth Sreenivasa kmem_free(smobj_array, array_bytes); 27435cabbc6bSPrashanth Sreenivasa VERIFY0(dmu_object_free(mos, vd->vdev_ms_array, tx)); 27445cabbc6bSPrashanth Sreenivasa vd->vdev_ms_array = 0; 27455cabbc6bSPrashanth Sreenivasa } 27465cabbc6bSPrashanth Sreenivasa 27475cabbc6bSPrashanth Sreenivasa static void 27485cabbc6bSPrashanth Sreenivasa vdev_remove_empty(vdev_t *vd, uint64_t txg) 274988ecc943SGeorge Wilson { 275088ecc943SGeorge Wilson spa_t *spa = vd->vdev_spa; 275188ecc943SGeorge Wilson dmu_tx_t *tx; 275288ecc943SGeorge Wilson 2753215198a6SJoe Stein ASSERT(vd == vd->vdev_top); 2754215198a6SJoe Stein ASSERT3U(txg, ==, spa_syncing_txg(spa)); 275588ecc943SGeorge Wilson 275688ecc943SGeorge Wilson if (vd->vdev_ms != NULL) { 27572e4c9986SGeorge Wilson metaslab_group_t *mg = vd->vdev_mg; 27582e4c9986SGeorge Wilson 27592e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 27602e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 27612e4c9986SGeorge Wilson 276288ecc943SGeorge Wilson for (int m = 0; m < vd->vdev_ms_count; m++) { 276388ecc943SGeorge Wilson metaslab_t *msp = vd->vdev_ms[m]; 276488ecc943SGeorge Wilson 27650713e232SGeorge Wilson if (msp == NULL || msp->ms_sm == NULL) 276688ecc943SGeorge Wilson continue; 276788ecc943SGeorge Wilson 27680713e232SGeorge Wilson mutex_enter(&msp->ms_lock); 27692e4c9986SGeorge Wilson /* 27702e4c9986SGeorge Wilson * If the metaslab was not loaded when the vdev 27712e4c9986SGeorge Wilson * was removed then the histogram accounting may 27722e4c9986SGeorge Wilson * not be accurate. Update the histogram information 27732e4c9986SGeorge Wilson * here so that we ensure that the metaslab group 27742e4c9986SGeorge Wilson * and metaslab class are up-to-date. 27752e4c9986SGeorge Wilson */ 27762e4c9986SGeorge Wilson metaslab_group_histogram_remove(mg, msp); 27772e4c9986SGeorge Wilson 27780713e232SGeorge Wilson VERIFY0(space_map_allocated(msp->ms_sm)); 27790713e232SGeorge Wilson space_map_close(msp->ms_sm); 27800713e232SGeorge Wilson msp->ms_sm = NULL; 27810713e232SGeorge Wilson mutex_exit(&msp->ms_lock); 278288ecc943SGeorge Wilson } 27832e4c9986SGeorge Wilson 2784*86714001SSerapheim Dimitropoulos if (vd->vdev_checkpoint_sm != NULL) { 2785*86714001SSerapheim Dimitropoulos ASSERT(spa_has_checkpoint(spa)); 2786*86714001SSerapheim Dimitropoulos space_map_close(vd->vdev_checkpoint_sm); 2787*86714001SSerapheim Dimitropoulos vd->vdev_checkpoint_sm = NULL; 2788*86714001SSerapheim Dimitropoulos } 2789*86714001SSerapheim Dimitropoulos 27902e4c9986SGeorge Wilson metaslab_group_histogram_verify(mg); 27912e4c9986SGeorge Wilson metaslab_class_histogram_verify(mg->mg_class); 27922e4c9986SGeorge Wilson for (int i = 0; i < RANGE_TREE_HISTOGRAM_SIZE; i++) 27932e4c9986SGeorge Wilson ASSERT0(mg->mg_histogram[i]); 279488ecc943SGeorge Wilson } 279588ecc943SGeorge Wilson 27965cabbc6bSPrashanth Sreenivasa tx = dmu_tx_create_assigned(spa_get_dsl(spa), txg); 27975cabbc6bSPrashanth Sreenivasa vdev_destroy_spacemaps(vd, tx); 2798215198a6SJoe Stein 2799215198a6SJoe Stein if (vd->vdev_islog && vd->vdev_top_zap != 0) { 2800215198a6SJoe Stein vdev_destroy_unlink_zap(vd, vd->vdev_top_zap, tx); 2801215198a6SJoe Stein vd->vdev_top_zap = 0; 2802215198a6SJoe Stein } 280388ecc943SGeorge Wilson dmu_tx_commit(tx); 280488ecc943SGeorge Wilson } 280588ecc943SGeorge Wilson 2806fa9e4066Sahrens void 2807fa9e4066Sahrens vdev_sync_done(vdev_t *vd, uint64_t txg) 2808fa9e4066Sahrens { 2809fa9e4066Sahrens metaslab_t *msp; 281080eb36f2SGeorge Wilson boolean_t reassess = !txg_list_empty(&vd->vdev_ms_list, TXG_CLEAN(txg)); 2811fa9e4066Sahrens 28125cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 281388ecc943SGeorge Wilson 2814fa9e4066Sahrens while (msp = txg_list_remove(&vd->vdev_ms_list, TXG_CLEAN(txg))) 2815fa9e4066Sahrens metaslab_sync_done(msp, txg); 281680eb36f2SGeorge Wilson 281780eb36f2SGeorge Wilson if (reassess) 281880eb36f2SGeorge Wilson metaslab_sync_reassess(vd->vdev_mg); 2819fa9e4066Sahrens } 2820fa9e4066Sahrens 2821fa9e4066Sahrens void 2822fa9e4066Sahrens vdev_sync(vdev_t *vd, uint64_t txg) 2823fa9e4066Sahrens { 2824fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 2825fa9e4066Sahrens vdev_t *lvd; 2826fa9e4066Sahrens metaslab_t *msp; 2827ecc2d604Sbonwick dmu_tx_t *tx; 2828fa9e4066Sahrens 28295cabbc6bSPrashanth Sreenivasa if (range_tree_space(vd->vdev_obsolete_segments) > 0) { 28305cabbc6bSPrashanth Sreenivasa dmu_tx_t *tx; 28315cabbc6bSPrashanth Sreenivasa 28325cabbc6bSPrashanth Sreenivasa ASSERT(vd->vdev_removing || 28335cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_indirect_ops); 283488ecc943SGeorge Wilson 28355cabbc6bSPrashanth Sreenivasa tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 28365cabbc6bSPrashanth Sreenivasa vdev_indirect_sync_obsolete(vd, tx); 28375cabbc6bSPrashanth Sreenivasa dmu_tx_commit(tx); 28385cabbc6bSPrashanth Sreenivasa 28395cabbc6bSPrashanth Sreenivasa /* 28405cabbc6bSPrashanth Sreenivasa * If the vdev is indirect, it can't have dirty 28415cabbc6bSPrashanth Sreenivasa * metaslabs or DTLs. 28425cabbc6bSPrashanth Sreenivasa */ 28435cabbc6bSPrashanth Sreenivasa if (vd->vdev_ops == &vdev_indirect_ops) { 28445cabbc6bSPrashanth Sreenivasa ASSERT(txg_list_empty(&vd->vdev_ms_list, txg)); 28455cabbc6bSPrashanth Sreenivasa ASSERT(txg_list_empty(&vd->vdev_dtl_list, txg)); 28465cabbc6bSPrashanth Sreenivasa return; 28475cabbc6bSPrashanth Sreenivasa } 28485cabbc6bSPrashanth Sreenivasa } 28495cabbc6bSPrashanth Sreenivasa 28505cabbc6bSPrashanth Sreenivasa ASSERT(vdev_is_concrete(vd)); 28515cabbc6bSPrashanth Sreenivasa 28525cabbc6bSPrashanth Sreenivasa if (vd->vdev_ms_array == 0 && vd->vdev_ms_shift != 0 && 28535cabbc6bSPrashanth Sreenivasa !vd->vdev_removing) { 2854ecc2d604Sbonwick ASSERT(vd == vd->vdev_top); 28555cabbc6bSPrashanth Sreenivasa ASSERT0(vd->vdev_indirect_config.vic_mapping_object); 2856ecc2d604Sbonwick tx = dmu_tx_create_assigned(spa->spa_dsl_pool, txg); 2857ecc2d604Sbonwick vd->vdev_ms_array = dmu_object_alloc(spa->spa_meta_objset, 2858ecc2d604Sbonwick DMU_OT_OBJECT_ARRAY, 0, DMU_OT_NONE, 0, tx); 2859ecc2d604Sbonwick ASSERT(vd->vdev_ms_array != 0); 2860ecc2d604Sbonwick vdev_config_dirty(vd); 2861ecc2d604Sbonwick dmu_tx_commit(tx); 2862ecc2d604Sbonwick } 2863fa9e4066Sahrens 2864ecc2d604Sbonwick while ((msp = txg_list_remove(&vd->vdev_ms_list, txg)) != NULL) { 2865fa9e4066Sahrens metaslab_sync(msp, txg); 2866ecc2d604Sbonwick (void) txg_list_add(&vd->vdev_ms_list, msp, TXG_CLEAN(txg)); 2867ecc2d604Sbonwick } 2868fa9e4066Sahrens 2869fa9e4066Sahrens while ((lvd = txg_list_remove(&vd->vdev_dtl_list, txg)) != NULL) 2870fa9e4066Sahrens vdev_dtl_sync(lvd, txg); 2871fa9e4066Sahrens 28725cabbc6bSPrashanth Sreenivasa /* 28735cabbc6bSPrashanth Sreenivasa * Remove the metadata associated with this vdev once it's empty. 28745cabbc6bSPrashanth Sreenivasa * Note that this is typically used for log/cache device removal; 28755cabbc6bSPrashanth Sreenivasa * we don't empty toplevel vdevs when removing them. But if 28765cabbc6bSPrashanth Sreenivasa * a toplevel happens to be emptied, this is not harmful. 28775cabbc6bSPrashanth Sreenivasa */ 28785cabbc6bSPrashanth Sreenivasa if (vd->vdev_stat.vs_alloc == 0 && vd->vdev_removing) { 28795cabbc6bSPrashanth Sreenivasa vdev_remove_empty(vd, txg); 28805cabbc6bSPrashanth Sreenivasa } 28815cabbc6bSPrashanth Sreenivasa 2882fa9e4066Sahrens (void) txg_list_add(&spa->spa_vdev_txg_list, vd, TXG_CLEAN(txg)); 2883fa9e4066Sahrens } 2884fa9e4066Sahrens 2885fa9e4066Sahrens uint64_t 2886fa9e4066Sahrens vdev_psize_to_asize(vdev_t *vd, uint64_t psize) 2887fa9e4066Sahrens { 2888fa9e4066Sahrens return (vd->vdev_ops->vdev_op_asize(vd, psize)); 2889fa9e4066Sahrens } 2890fa9e4066Sahrens 28913d7072f8Seschrock /* 28923d7072f8Seschrock * Mark the given vdev faulted. A faulted vdev behaves as if the device could 28933d7072f8Seschrock * not be opened, and no I/O is attempted. 28943d7072f8Seschrock */ 2895fa9e4066Sahrens int 2896069f55e2SEric Schrock vdev_fault(spa_t *spa, uint64_t guid, vdev_aux_t aux) 2897fa9e4066Sahrens { 28984b964adaSGeorge Wilson vdev_t *vd, *tvd; 2899fa9e4066Sahrens 29008f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 2901fa9e4066Sahrens 2902c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2903e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2904e14bb325SJeff Bonwick 29053d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2906e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 2907fa9e4066Sahrens 29084b964adaSGeorge Wilson tvd = vd->vdev_top; 29094b964adaSGeorge Wilson 2910069f55e2SEric Schrock /* 2911069f55e2SEric Schrock * We don't directly use the aux state here, but if we do a 2912069f55e2SEric Schrock * vdev_reopen(), we need this value to be present to remember why we 2913069f55e2SEric Schrock * were faulted. 2914069f55e2SEric Schrock */ 2915069f55e2SEric Schrock vd->vdev_label_aux = aux; 2916069f55e2SEric Schrock 29173d7072f8Seschrock /* 29183d7072f8Seschrock * Faulted state takes precedence over degraded. 29193d7072f8Seschrock */ 292098d1cbfeSGeorge Wilson vd->vdev_delayed_close = B_FALSE; 29213d7072f8Seschrock vd->vdev_faulted = 1ULL; 29223d7072f8Seschrock vd->vdev_degraded = 0ULL; 2923069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_FAULTED, aux); 29243d7072f8Seschrock 29253d7072f8Seschrock /* 2926c79790bcSGeorge Wilson * If this device has the only valid copy of the data, then 2927c79790bcSGeorge Wilson * back off and simply mark the vdev as degraded instead. 29283d7072f8Seschrock */ 29294b964adaSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && vdev_dtl_required(vd)) { 29303d7072f8Seschrock vd->vdev_degraded = 1ULL; 29313d7072f8Seschrock vd->vdev_faulted = 0ULL; 29323d7072f8Seschrock 29333d7072f8Seschrock /* 29343d7072f8Seschrock * If we reopen the device and it's not dead, only then do we 29353d7072f8Seschrock * mark it degraded. 29363d7072f8Seschrock */ 29374b964adaSGeorge Wilson vdev_reopen(tvd); 29383d7072f8Seschrock 2939069f55e2SEric Schrock if (vdev_readable(vd)) 2940069f55e2SEric Schrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, aux); 29413d7072f8Seschrock } 29423d7072f8Seschrock 2943e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 29443d7072f8Seschrock } 29453d7072f8Seschrock 29463d7072f8Seschrock /* 29473d7072f8Seschrock * Mark the given vdev degraded. A degraded vdev is purely an indication to the 29483d7072f8Seschrock * user that something is wrong. The vdev continues to operate as normal as far 29493d7072f8Seschrock * as I/O is concerned. 29503d7072f8Seschrock */ 29513d7072f8Seschrock int 2952069f55e2SEric Schrock vdev_degrade(spa_t *spa, uint64_t guid, vdev_aux_t aux) 29533d7072f8Seschrock { 2954c5904d13Seschrock vdev_t *vd; 29550a4e9518Sgw 29568f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 29573d7072f8Seschrock 2958c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2959e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 2960e14bb325SJeff Bonwick 29610e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 2962e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 29630e34b6a7Sbonwick 29643d7072f8Seschrock /* 29653d7072f8Seschrock * If the vdev is already faulted, then don't do anything. 29663d7072f8Seschrock */ 2967e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded) 2968e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, 0)); 29693d7072f8Seschrock 29703d7072f8Seschrock vd->vdev_degraded = 1ULL; 29713d7072f8Seschrock if (!vdev_is_dead(vd)) 29723d7072f8Seschrock vdev_set_state(vd, B_FALSE, VDEV_STATE_DEGRADED, 2973069f55e2SEric Schrock aux); 29743d7072f8Seschrock 2975e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 29763d7072f8Seschrock } 29773d7072f8Seschrock 29783d7072f8Seschrock /* 2979f7170741SWill Andrews * Online the given vdev. 2980f7170741SWill Andrews * 2981f7170741SWill Andrews * If 'ZFS_ONLINE_UNSPARE' is set, it implies two things. First, any attached 2982f7170741SWill Andrews * spare device should be detached when the device finishes resilvering. 2983f7170741SWill Andrews * Second, the online should be treated like a 'test' online case, so no FMA 2984f7170741SWill Andrews * events are generated if the device fails to open. 29853d7072f8Seschrock */ 29863d7072f8Seschrock int 2987e14bb325SJeff Bonwick vdev_online(spa_t *spa, uint64_t guid, uint64_t flags, vdev_state_t *newstate) 29883d7072f8Seschrock { 2989573ca77eSGeorge Wilson vdev_t *vd, *tvd, *pvd, *rvd = spa->spa_root_vdev; 29905f368aefSYuri Pankov boolean_t wasoffline; 29915f368aefSYuri Pankov vdev_state_t oldstate; 29923d7072f8Seschrock 29938f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_NONE); 29943d7072f8Seschrock 2995c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 2996e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 29973d7072f8Seschrock 29983d7072f8Seschrock if (!vd->vdev_ops->vdev_op_leaf) 2999e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 3000fa9e4066Sahrens 30015f368aefSYuri Pankov wasoffline = (vd->vdev_offline || vd->vdev_tmpoffline); 30025f368aefSYuri Pankov oldstate = vd->vdev_state; 300314372834SHans Rosenfeld 3004573ca77eSGeorge Wilson tvd = vd->vdev_top; 3005fa9e4066Sahrens vd->vdev_offline = B_FALSE; 3006441d80aaSlling vd->vdev_tmpoffline = B_FALSE; 3007e14bb325SJeff Bonwick vd->vdev_checkremove = !!(flags & ZFS_ONLINE_CHECKREMOVE); 3008e14bb325SJeff Bonwick vd->vdev_forcefault = !!(flags & ZFS_ONLINE_FORCEFAULT); 3009573ca77eSGeorge Wilson 3010573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 3011573ca77eSGeorge Wilson if (!vd->vdev_aux) { 3012573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3013573ca77eSGeorge Wilson pvd->vdev_expanding = !!(flags & ZFS_ONLINE_EXPAND); 3014573ca77eSGeorge Wilson } 3015573ca77eSGeorge Wilson 3016573ca77eSGeorge Wilson vdev_reopen(tvd); 30173d7072f8Seschrock vd->vdev_checkremove = vd->vdev_forcefault = B_FALSE; 30183d7072f8Seschrock 3019573ca77eSGeorge Wilson if (!vd->vdev_aux) { 3020573ca77eSGeorge Wilson for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 3021573ca77eSGeorge Wilson pvd->vdev_expanding = B_FALSE; 3022573ca77eSGeorge Wilson } 3023573ca77eSGeorge Wilson 30243d7072f8Seschrock if (newstate) 30253d7072f8Seschrock *newstate = vd->vdev_state; 30263d7072f8Seschrock if ((flags & ZFS_ONLINE_UNSPARE) && 30273d7072f8Seschrock !vdev_is_dead(vd) && vd->vdev_parent && 30283d7072f8Seschrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 30293d7072f8Seschrock vd->vdev_parent->vdev_child[0] == vd) 30303d7072f8Seschrock vd->vdev_unspare = B_TRUE; 3031fa9e4066Sahrens 3032573ca77eSGeorge Wilson if ((flags & ZFS_ONLINE_EXPAND) || spa->spa_autoexpand) { 3033573ca77eSGeorge Wilson 3034573ca77eSGeorge Wilson /* XXX - L2ARC 1.0 does not support expansion */ 3035573ca77eSGeorge Wilson if (vd->vdev_aux) 3036573ca77eSGeorge Wilson return (spa_vdev_state_exit(spa, vd, ENOTSUP)); 3037573ca77eSGeorge Wilson spa_async_request(spa, SPA_ASYNC_CONFIG_UPDATE); 3038573ca77eSGeorge Wilson } 303914372834SHans Rosenfeld 30405f368aefSYuri Pankov if (wasoffline || 30415f368aefSYuri Pankov (oldstate < VDEV_STATE_DEGRADED && 30425f368aefSYuri Pankov vd->vdev_state >= VDEV_STATE_DEGRADED)) 3043ce1577b0SDave Eddy spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_ONLINE); 304414372834SHans Rosenfeld 30458ad4d6ddSJeff Bonwick return (spa_vdev_state_exit(spa, vd, 0)); 3046fa9e4066Sahrens } 3047fa9e4066Sahrens 3048a1521560SJeff Bonwick static int 3049a1521560SJeff Bonwick vdev_offline_locked(spa_t *spa, uint64_t guid, uint64_t flags) 3050fa9e4066Sahrens { 3051e6ca193dSGeorge Wilson vdev_t *vd, *tvd; 30528f18d1faSGeorge Wilson int error = 0; 30538f18d1faSGeorge Wilson uint64_t generation; 30548f18d1faSGeorge Wilson metaslab_group_t *mg; 30550a4e9518Sgw 30568f18d1faSGeorge Wilson top: 30578f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 3058fa9e4066Sahrens 3059c5904d13Seschrock if ((vd = spa_lookup_by_guid(spa, guid, B_TRUE)) == NULL) 3060e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENODEV)); 3061fa9e4066Sahrens 30620e34b6a7Sbonwick if (!vd->vdev_ops->vdev_op_leaf) 3063e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, ENOTSUP)); 30640e34b6a7Sbonwick 3065e6ca193dSGeorge Wilson tvd = vd->vdev_top; 30668f18d1faSGeorge Wilson mg = tvd->vdev_mg; 30678f18d1faSGeorge Wilson generation = spa->spa_config_generation + 1; 3068e6ca193dSGeorge Wilson 3069fa9e4066Sahrens /* 3070ecc2d604Sbonwick * If the device isn't already offline, try to offline it. 3071fa9e4066Sahrens */ 3072ecc2d604Sbonwick if (!vd->vdev_offline) { 3073ecc2d604Sbonwick /* 30748ad4d6ddSJeff Bonwick * If this device has the only valid copy of some data, 3075e6ca193dSGeorge Wilson * don't allow it to be offlined. Log devices are always 3076e6ca193dSGeorge Wilson * expendable. 3077ecc2d604Sbonwick */ 3078e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3079e6ca193dSGeorge Wilson vdev_dtl_required(vd)) 3080e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3081fa9e4066Sahrens 30828f18d1faSGeorge Wilson /* 3083b24ab676SJeff Bonwick * If the top-level is a slog and it has had allocations 3084b24ab676SJeff Bonwick * then proceed. We check that the vdev's metaslab group 3085b24ab676SJeff Bonwick * is not NULL since it's possible that we may have just 3086b24ab676SJeff Bonwick * added this vdev but not yet initialized its metaslabs. 30878f18d1faSGeorge Wilson */ 30888f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) { 30898f18d1faSGeorge Wilson /* 30908f18d1faSGeorge Wilson * Prevent any future allocations. 30918f18d1faSGeorge Wilson */ 3092a1521560SJeff Bonwick metaslab_group_passivate(mg); 30938f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 30948f18d1faSGeorge Wilson 30955cabbc6bSPrashanth Sreenivasa error = spa_reset_logs(spa); 30968f18d1faSGeorge Wilson 3097*86714001SSerapheim Dimitropoulos /* 3098*86714001SSerapheim Dimitropoulos * If the log device was successfully reset but has 3099*86714001SSerapheim Dimitropoulos * checkpointed data, do not offline it. 3100*86714001SSerapheim Dimitropoulos */ 3101*86714001SSerapheim Dimitropoulos if (error == 0 && 3102*86714001SSerapheim Dimitropoulos tvd->vdev_checkpoint_sm != NULL) { 3103*86714001SSerapheim Dimitropoulos ASSERT3U(tvd->vdev_checkpoint_sm->sm_alloc, 3104*86714001SSerapheim Dimitropoulos !=, 0); 3105*86714001SSerapheim Dimitropoulos error = ZFS_ERR_CHECKPOINT_EXISTS; 3106*86714001SSerapheim Dimitropoulos } 3107*86714001SSerapheim Dimitropoulos 31088f18d1faSGeorge Wilson spa_vdev_state_enter(spa, SCL_ALLOC); 31098f18d1faSGeorge Wilson 31108f18d1faSGeorge Wilson /* 31118f18d1faSGeorge Wilson * Check to see if the config has changed. 31128f18d1faSGeorge Wilson */ 31138f18d1faSGeorge Wilson if (error || generation != spa->spa_config_generation) { 3114a1521560SJeff Bonwick metaslab_group_activate(mg); 31158f18d1faSGeorge Wilson if (error) 31168f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, 31178f18d1faSGeorge Wilson vd, error)); 31188f18d1faSGeorge Wilson (void) spa_vdev_state_exit(spa, vd, 0); 31198f18d1faSGeorge Wilson goto top; 31208f18d1faSGeorge Wilson } 3121fb09f5aaSMadhav Suresh ASSERT0(tvd->vdev_stat.vs_alloc); 31228f18d1faSGeorge Wilson } 31238f18d1faSGeorge Wilson 3124ecc2d604Sbonwick /* 3125ecc2d604Sbonwick * Offline this device and reopen its top-level vdev. 3126e6ca193dSGeorge Wilson * If the top-level vdev is a log device then just offline 3127e6ca193dSGeorge Wilson * it. Otherwise, if this action results in the top-level 3128e6ca193dSGeorge Wilson * vdev becoming unusable, undo it and fail the request. 3129ecc2d604Sbonwick */ 3130ecc2d604Sbonwick vd->vdev_offline = B_TRUE; 3131e6ca193dSGeorge Wilson vdev_reopen(tvd); 3132e6ca193dSGeorge Wilson 3133e6ca193dSGeorge Wilson if (!tvd->vdev_islog && vd->vdev_aux == NULL && 3134e6ca193dSGeorge Wilson vdev_is_dead(tvd)) { 3135ecc2d604Sbonwick vd->vdev_offline = B_FALSE; 3136e6ca193dSGeorge Wilson vdev_reopen(tvd); 3137e14bb325SJeff Bonwick return (spa_vdev_state_exit(spa, NULL, EBUSY)); 3138ecc2d604Sbonwick } 31398f18d1faSGeorge Wilson 31408f18d1faSGeorge Wilson /* 31418f18d1faSGeorge Wilson * Add the device back into the metaslab rotor so that 31428f18d1faSGeorge Wilson * once we online the device it's open for business. 31438f18d1faSGeorge Wilson */ 31448f18d1faSGeorge Wilson if (tvd->vdev_islog && mg != NULL) 3145a1521560SJeff Bonwick metaslab_group_activate(mg); 3146fa9e4066Sahrens } 3147fa9e4066Sahrens 3148e14bb325SJeff Bonwick vd->vdev_tmpoffline = !!(flags & ZFS_OFFLINE_TEMPORARY); 3149ecc2d604Sbonwick 31508f18d1faSGeorge Wilson return (spa_vdev_state_exit(spa, vd, 0)); 3151fa9e4066Sahrens } 3152fa9e4066Sahrens 3153a1521560SJeff Bonwick int 3154a1521560SJeff Bonwick vdev_offline(spa_t *spa, uint64_t guid, uint64_t flags) 3155a1521560SJeff Bonwick { 3156a1521560SJeff Bonwick int error; 3157a1521560SJeff Bonwick 3158a1521560SJeff Bonwick mutex_enter(&spa->spa_vdev_top_lock); 3159a1521560SJeff Bonwick error = vdev_offline_locked(spa, guid, flags); 3160a1521560SJeff Bonwick mutex_exit(&spa->spa_vdev_top_lock); 3161a1521560SJeff Bonwick 3162a1521560SJeff Bonwick return (error); 3163a1521560SJeff Bonwick } 3164a1521560SJeff Bonwick 3165ea8dc4b6Seschrock /* 3166ea8dc4b6Seschrock * Clear the error counts associated with this vdev. Unlike vdev_online() and 3167ea8dc4b6Seschrock * vdev_offline(), we assume the spa config is locked. We also clear all 3168ea8dc4b6Seschrock * children. If 'vd' is NULL, then the user wants to clear all vdevs. 3169ea8dc4b6Seschrock */ 3170ea8dc4b6Seschrock void 3171e14bb325SJeff Bonwick vdev_clear(spa_t *spa, vdev_t *vd) 3172fa9e4066Sahrens { 3173e14bb325SJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 3174e14bb325SJeff Bonwick 3175e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE_ALL, RW_WRITER) == SCL_STATE_ALL); 3176fa9e4066Sahrens 3177ea8dc4b6Seschrock if (vd == NULL) 3178e14bb325SJeff Bonwick vd = rvd; 3179fa9e4066Sahrens 3180ea8dc4b6Seschrock vd->vdev_stat.vs_read_errors = 0; 3181ea8dc4b6Seschrock vd->vdev_stat.vs_write_errors = 0; 3182ea8dc4b6Seschrock vd->vdev_stat.vs_checksum_errors = 0; 3183fa9e4066Sahrens 3184e14bb325SJeff Bonwick for (int c = 0; c < vd->vdev_children; c++) 3185e14bb325SJeff Bonwick vdev_clear(spa, vd->vdev_child[c]); 31863d7072f8Seschrock 31875cabbc6bSPrashanth Sreenivasa /* 31885cabbc6bSPrashanth Sreenivasa * It makes no sense to "clear" an indirect vdev. 31895cabbc6bSPrashanth Sreenivasa */ 31905cabbc6bSPrashanth Sreenivasa if (!vdev_is_concrete(vd)) 31915cabbc6bSPrashanth Sreenivasa return; 31925cabbc6bSPrashanth Sreenivasa 31933d7072f8Seschrock /* 31948a79c1b5Sek * If we're in the FAULTED state or have experienced failed I/O, then 31958a79c1b5Sek * clear the persistent state and attempt to reopen the device. We 31968a79c1b5Sek * also mark the vdev config dirty, so that the new faulted state is 31978a79c1b5Sek * written out to disk. 31983d7072f8Seschrock */ 3199e14bb325SJeff Bonwick if (vd->vdev_faulted || vd->vdev_degraded || 3200e14bb325SJeff Bonwick !vdev_readable(vd) || !vdev_writeable(vd)) { 32018a79c1b5Sek 3202096d22d4SEric Schrock /* 3203096d22d4SEric Schrock * When reopening in reponse to a clear event, it may be due to 3204096d22d4SEric Schrock * a fmadm repair request. In this case, if the device is 3205096d22d4SEric Schrock * still broken, we want to still post the ereport again. 3206096d22d4SEric Schrock */ 3207096d22d4SEric Schrock vd->vdev_forcefault = B_TRUE; 3208096d22d4SEric Schrock 32094b964adaSGeorge Wilson vd->vdev_faulted = vd->vdev_degraded = 0ULL; 3210e14bb325SJeff Bonwick vd->vdev_cant_read = B_FALSE; 3211e14bb325SJeff Bonwick vd->vdev_cant_write = B_FALSE; 3212e14bb325SJeff Bonwick 3213f9af39baSGeorge Wilson vdev_reopen(vd == rvd ? rvd : vd->vdev_top); 32143d7072f8Seschrock 3215096d22d4SEric Schrock vd->vdev_forcefault = B_FALSE; 3216096d22d4SEric Schrock 3217f9af39baSGeorge Wilson if (vd != rvd && vdev_writeable(vd->vdev_top)) 3218e14bb325SJeff Bonwick vdev_state_dirty(vd->vdev_top); 3219e14bb325SJeff Bonwick 3220e14bb325SJeff Bonwick if (vd->vdev_aux == NULL && !vdev_is_dead(vd)) 3221bb8b5132Sek spa_async_request(spa, SPA_ASYNC_RESILVER); 32223d7072f8Seschrock 3223ce1577b0SDave Eddy spa_event_notify(spa, vd, NULL, ESC_ZFS_VDEV_CLEAR); 32243d7072f8Seschrock } 3225096d22d4SEric Schrock 3226096d22d4SEric Schrock /* 3227096d22d4SEric Schrock * When clearing a FMA-diagnosed fault, we always want to 3228096d22d4SEric Schrock * unspare the device, as we assume that the original spare was 3229096d22d4SEric Schrock * done in response to the FMA fault. 3230096d22d4SEric Schrock */ 3231096d22d4SEric Schrock if (!vdev_is_dead(vd) && vd->vdev_parent != NULL && 3232096d22d4SEric Schrock vd->vdev_parent->vdev_ops == &vdev_spare_ops && 3233096d22d4SEric Schrock vd->vdev_parent->vdev_child[0] == vd) 3234096d22d4SEric Schrock vd->vdev_unspare = B_TRUE; 3235fa9e4066Sahrens } 3236fa9e4066Sahrens 3237e14bb325SJeff Bonwick boolean_t 3238e14bb325SJeff Bonwick vdev_is_dead(vdev_t *vd) 32390a4e9518Sgw { 324088ecc943SGeorge Wilson /* 324188ecc943SGeorge Wilson * Holes and missing devices are always considered "dead". 324288ecc943SGeorge Wilson * This simplifies the code since we don't have to check for 324388ecc943SGeorge Wilson * these types of devices in the various code paths. 324488ecc943SGeorge Wilson * Instead we rely on the fact that we skip over dead devices 324588ecc943SGeorge Wilson * before issuing I/O to them. 324688ecc943SGeorge Wilson */ 32475cabbc6bSPrashanth Sreenivasa return (vd->vdev_state < VDEV_STATE_DEGRADED || 32485cabbc6bSPrashanth Sreenivasa vd->vdev_ops == &vdev_hole_ops || 324988ecc943SGeorge Wilson vd->vdev_ops == &vdev_missing_ops); 32500a4e9518Sgw } 32510a4e9518Sgw 3252e14bb325SJeff Bonwick boolean_t 3253e14bb325SJeff Bonwick vdev_readable(vdev_t *vd) 32540a4e9518Sgw { 3255e14bb325SJeff Bonwick return (!vdev_is_dead(vd) && !vd->vdev_cant_read); 32560a4e9518Sgw } 32570a4e9518Sgw 3258e14bb325SJeff Bonwick boolean_t 3259e14bb325SJeff Bonwick vdev_writeable(vdev_t *vd) 3260fa9e4066Sahrens { 32615cabbc6bSPrashanth Sreenivasa return (!vdev_is_dead(vd) && !vd->vdev_cant_write && 32625cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)); 3263fa9e4066Sahrens } 3264fa9e4066Sahrens 3265a31e6787SGeorge Wilson boolean_t 3266a31e6787SGeorge Wilson vdev_allocatable(vdev_t *vd) 3267a31e6787SGeorge Wilson { 32688ad4d6ddSJeff Bonwick uint64_t state = vd->vdev_state; 32698ad4d6ddSJeff Bonwick 3270a31e6787SGeorge Wilson /* 32718ad4d6ddSJeff Bonwick * We currently allow allocations from vdevs which may be in the 3272a31e6787SGeorge Wilson * process of reopening (i.e. VDEV_STATE_CLOSED). If the device 3273a31e6787SGeorge Wilson * fails to reopen then we'll catch it later when we're holding 32748ad4d6ddSJeff Bonwick * the proper locks. Note that we have to get the vdev state 32758ad4d6ddSJeff Bonwick * in a local variable because although it changes atomically, 32768ad4d6ddSJeff Bonwick * we're asking two separate questions about it. 3277a31e6787SGeorge Wilson */ 32788ad4d6ddSJeff Bonwick return (!(state < VDEV_STATE_DEGRADED && state != VDEV_STATE_CLOSED) && 32795cabbc6bSPrashanth Sreenivasa !vd->vdev_cant_write && vdev_is_concrete(vd) && 32800f7643c7SGeorge Wilson vd->vdev_mg->mg_initialized); 3281a31e6787SGeorge Wilson } 3282a31e6787SGeorge Wilson 3283e14bb325SJeff Bonwick boolean_t 3284e14bb325SJeff Bonwick vdev_accessible(vdev_t *vd, zio_t *zio) 3285fa9e4066Sahrens { 3286e14bb325SJeff Bonwick ASSERT(zio->io_vd == vd); 3287fa9e4066Sahrens 3288e14bb325SJeff Bonwick if (vdev_is_dead(vd) || vd->vdev_remove_wanted) 3289e14bb325SJeff Bonwick return (B_FALSE); 3290fa9e4066Sahrens 3291e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_READ) 3292e14bb325SJeff Bonwick return (!vd->vdev_cant_read); 3293fa9e4066Sahrens 3294e14bb325SJeff Bonwick if (zio->io_type == ZIO_TYPE_WRITE) 3295e14bb325SJeff Bonwick return (!vd->vdev_cant_write); 3296fa9e4066Sahrens 3297e14bb325SJeff Bonwick return (B_TRUE); 3298fa9e4066Sahrens } 3299fa9e4066Sahrens 3300*86714001SSerapheim Dimitropoulos boolean_t 3301*86714001SSerapheim Dimitropoulos vdev_is_spacemap_addressable(vdev_t *vd) 3302*86714001SSerapheim Dimitropoulos { 3303*86714001SSerapheim Dimitropoulos /* 3304*86714001SSerapheim Dimitropoulos * Assuming 47 bits of the space map entry dedicated for the entry's 3305*86714001SSerapheim Dimitropoulos * offset (see description in space_map.h), we calculate the maximum 3306*86714001SSerapheim Dimitropoulos * address that can be described by a space map entry for the given 3307*86714001SSerapheim Dimitropoulos * device. 3308*86714001SSerapheim Dimitropoulos */ 3309*86714001SSerapheim Dimitropoulos uint64_t shift = vd->vdev_ashift + 47; 3310*86714001SSerapheim Dimitropoulos 3311*86714001SSerapheim Dimitropoulos if (shift >= 63) /* detect potential overflow */ 3312*86714001SSerapheim Dimitropoulos return (B_TRUE); 3313*86714001SSerapheim Dimitropoulos 3314*86714001SSerapheim Dimitropoulos return (vd->vdev_asize < (1ULL << shift)); 3315*86714001SSerapheim Dimitropoulos } 3316*86714001SSerapheim Dimitropoulos 3317fa9e4066Sahrens /* 3318fa9e4066Sahrens * Get statistics for the given vdev. 3319fa9e4066Sahrens */ 3320fa9e4066Sahrens void 3321fa9e4066Sahrens vdev_get_stats(vdev_t *vd, vdev_stat_t *vs) 3322fa9e4066Sahrens { 33232e4c9986SGeorge Wilson spa_t *spa = vd->vdev_spa; 33242e4c9986SGeorge Wilson vdev_t *rvd = spa->spa_root_vdev; 3325c39a2aaeSGeorge Wilson vdev_t *tvd = vd->vdev_top; 33262e4c9986SGeorge Wilson 33272e4c9986SGeorge Wilson ASSERT(spa_config_held(spa, SCL_ALL, RW_READER) != 0); 3328fa9e4066Sahrens 3329fa9e4066Sahrens mutex_enter(&vd->vdev_stat_lock); 3330fa9e4066Sahrens bcopy(&vd->vdev_stat, vs, sizeof (*vs)); 3331fa9e4066Sahrens vs->vs_timestamp = gethrtime() - vs->vs_timestamp; 3332fa9e4066Sahrens vs->vs_state = vd->vdev_state; 3333573ca77eSGeorge Wilson vs->vs_rsize = vdev_get_min_asize(vd); 3334573ca77eSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf) 3335573ca77eSGeorge Wilson vs->vs_rsize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE; 3336c39a2aaeSGeorge Wilson /* 3337c39a2aaeSGeorge Wilson * Report expandable space on top-level, non-auxillary devices only. 3338c39a2aaeSGeorge Wilson * The expandable space is reported in terms of metaslab sized units 3339c39a2aaeSGeorge Wilson * since that determines how much space the pool can expand. 3340c39a2aaeSGeorge Wilson */ 3341c39a2aaeSGeorge Wilson if (vd->vdev_aux == NULL && tvd != NULL) { 33427855d95bSToomas Soome vs->vs_esize = P2ALIGN(vd->vdev_max_asize - vd->vdev_asize - 33437855d95bSToomas Soome spa->spa_bootsize, 1ULL << tvd->vdev_ms_shift); 3344c39a2aaeSGeorge Wilson } 33455cabbc6bSPrashanth Sreenivasa if (vd->vdev_aux == NULL && vd == vd->vdev_top && 33465cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) { 33472e4c9986SGeorge Wilson vs->vs_fragmentation = vd->vdev_mg->mg_fragmentation; 33482986efa8SAlex Reece } 3349fa9e4066Sahrens 3350fa9e4066Sahrens /* 3351fa9e4066Sahrens * If we're getting stats on the root vdev, aggregate the I/O counts 3352fa9e4066Sahrens * over all top-level vdevs (i.e. the direct children of the root). 3353fa9e4066Sahrens */ 3354fa9e4066Sahrens if (vd == rvd) { 3355e14bb325SJeff Bonwick for (int c = 0; c < rvd->vdev_children; c++) { 3356fa9e4066Sahrens vdev_t *cvd = rvd->vdev_child[c]; 3357fa9e4066Sahrens vdev_stat_t *cvs = &cvd->vdev_stat; 3358fa9e4066Sahrens 3359e14bb325SJeff Bonwick for (int t = 0; t < ZIO_TYPES; t++) { 3360fa9e4066Sahrens vs->vs_ops[t] += cvs->vs_ops[t]; 3361fa9e4066Sahrens vs->vs_bytes[t] += cvs->vs_bytes[t]; 3362fa9e4066Sahrens } 33633f9d6ad7SLin Ling cvs->vs_scan_removing = cvd->vdev_removing; 3364fa9e4066Sahrens } 3365fa9e4066Sahrens } 33662e4c9986SGeorge Wilson mutex_exit(&vd->vdev_stat_lock); 3367fa9e4066Sahrens } 3368fa9e4066Sahrens 3369fa94a07fSbrendan void 3370fa94a07fSbrendan vdev_clear_stats(vdev_t *vd) 3371fa94a07fSbrendan { 3372fa94a07fSbrendan mutex_enter(&vd->vdev_stat_lock); 3373fa94a07fSbrendan vd->vdev_stat.vs_space = 0; 3374fa94a07fSbrendan vd->vdev_stat.vs_dspace = 0; 3375fa94a07fSbrendan vd->vdev_stat.vs_alloc = 0; 3376fa94a07fSbrendan mutex_exit(&vd->vdev_stat_lock); 3377fa94a07fSbrendan } 3378fa94a07fSbrendan 33793f9d6ad7SLin Ling void 33803f9d6ad7SLin Ling vdev_scan_stat_init(vdev_t *vd) 33813f9d6ad7SLin Ling { 33823f9d6ad7SLin Ling vdev_stat_t *vs = &vd->vdev_stat; 33833f9d6ad7SLin Ling 33843f9d6ad7SLin Ling for (int c = 0; c < vd->vdev_children; c++) 33853f9d6ad7SLin Ling vdev_scan_stat_init(vd->vdev_child[c]); 33863f9d6ad7SLin Ling 33873f9d6ad7SLin Ling mutex_enter(&vd->vdev_stat_lock); 33883f9d6ad7SLin Ling vs->vs_scan_processed = 0; 33893f9d6ad7SLin Ling mutex_exit(&vd->vdev_stat_lock); 33903f9d6ad7SLin Ling } 33913f9d6ad7SLin Ling 3392fa9e4066Sahrens void 3393e14bb325SJeff Bonwick vdev_stat_update(zio_t *zio, uint64_t psize) 3394fa9e4066Sahrens { 33958ad4d6ddSJeff Bonwick spa_t *spa = zio->io_spa; 33968ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 3397e14bb325SJeff Bonwick vdev_t *vd = zio->io_vd ? zio->io_vd : rvd; 3398fa9e4066Sahrens vdev_t *pvd; 3399fa9e4066Sahrens uint64_t txg = zio->io_txg; 3400fa9e4066Sahrens vdev_stat_t *vs = &vd->vdev_stat; 3401fa9e4066Sahrens zio_type_t type = zio->io_type; 3402fa9e4066Sahrens int flags = zio->io_flags; 3403fa9e4066Sahrens 3404e14bb325SJeff Bonwick /* 3405e14bb325SJeff Bonwick * If this i/o is a gang leader, it didn't do any actual work. 3406e14bb325SJeff Bonwick */ 3407e14bb325SJeff Bonwick if (zio->io_gang_tree) 3408e14bb325SJeff Bonwick return; 3409e14bb325SJeff Bonwick 3410fa9e4066Sahrens if (zio->io_error == 0) { 3411e14bb325SJeff Bonwick /* 3412e14bb325SJeff Bonwick * If this is a root i/o, don't count it -- we've already 3413e14bb325SJeff Bonwick * counted the top-level vdevs, and vdev_get_stats() will 3414e14bb325SJeff Bonwick * aggregate them when asked. This reduces contention on 3415e14bb325SJeff Bonwick * the root vdev_stat_lock and implicitly handles blocks 3416e14bb325SJeff Bonwick * that compress away to holes, for which there is no i/o. 3417e14bb325SJeff Bonwick * (Holes never create vdev children, so all the counters 3418e14bb325SJeff Bonwick * remain zero, which is what we want.) 3419e14bb325SJeff Bonwick * 3420e14bb325SJeff Bonwick * Note: this only applies to successful i/o (io_error == 0) 3421e14bb325SJeff Bonwick * because unlike i/o counts, errors are not additive. 3422e14bb325SJeff Bonwick * When reading a ditto block, for example, failure of 3423e14bb325SJeff Bonwick * one top-level vdev does not imply a root-level error. 3424e14bb325SJeff Bonwick */ 3425e14bb325SJeff Bonwick if (vd == rvd) 3426e14bb325SJeff Bonwick return; 3427e14bb325SJeff Bonwick 3428e14bb325SJeff Bonwick ASSERT(vd == zio->io_vd); 34298ad4d6ddSJeff Bonwick 34308ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_IO_BYPASS) 34318ad4d6ddSJeff Bonwick return; 34328ad4d6ddSJeff Bonwick 34338ad4d6ddSJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 34348ad4d6ddSJeff Bonwick 3435e14bb325SJeff Bonwick if (flags & ZIO_FLAG_IO_REPAIR) { 343644ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 34373f9d6ad7SLin Ling dsl_scan_phys_t *scn_phys = 34383f9d6ad7SLin Ling &spa->spa_dsl_pool->dp_scan->scn_phys; 34393f9d6ad7SLin Ling uint64_t *processed = &scn_phys->scn_processed; 34403f9d6ad7SLin Ling 34413f9d6ad7SLin Ling /* XXX cleanup? */ 34423f9d6ad7SLin Ling if (vd->vdev_ops->vdev_op_leaf) 34433f9d6ad7SLin Ling atomic_add_64(processed, psize); 34443f9d6ad7SLin Ling vs->vs_scan_processed += psize; 34453f9d6ad7SLin Ling } 34463f9d6ad7SLin Ling 34478ad4d6ddSJeff Bonwick if (flags & ZIO_FLAG_SELF_HEAL) 3448e14bb325SJeff Bonwick vs->vs_self_healed += psize; 3449fa9e4066Sahrens } 34508ad4d6ddSJeff Bonwick 34518ad4d6ddSJeff Bonwick vs->vs_ops[type]++; 34528ad4d6ddSJeff Bonwick vs->vs_bytes[type] += psize; 34538ad4d6ddSJeff Bonwick 34548ad4d6ddSJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 3455fa9e4066Sahrens return; 3456fa9e4066Sahrens } 3457fa9e4066Sahrens 3458fa9e4066Sahrens if (flags & ZIO_FLAG_SPECULATIVE) 3459fa9e4066Sahrens return; 3460fa9e4066Sahrens 34618956713aSEric Schrock /* 34628956713aSEric Schrock * If this is an I/O error that is going to be retried, then ignore the 34638956713aSEric Schrock * error. Otherwise, the user may interpret B_FAILFAST I/O errors as 34648956713aSEric Schrock * hard errors, when in reality they can happen for any number of 34658956713aSEric Schrock * innocuous reasons (bus resets, MPxIO link failure, etc). 34668956713aSEric Schrock */ 34678956713aSEric Schrock if (zio->io_error == EIO && 34688956713aSEric Schrock !(zio->io_flags & ZIO_FLAG_IO_RETRY)) 34698956713aSEric Schrock return; 34708956713aSEric Schrock 34718f18d1faSGeorge Wilson /* 34728f18d1faSGeorge Wilson * Intent logs writes won't propagate their error to the root 34738f18d1faSGeorge Wilson * I/O so don't mark these types of failures as pool-level 34748f18d1faSGeorge Wilson * errors. 34758f18d1faSGeorge Wilson */ 34768f18d1faSGeorge Wilson if (zio->io_vd == NULL && (zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) 34778f18d1faSGeorge Wilson return; 34788f18d1faSGeorge Wilson 3479e14bb325SJeff Bonwick mutex_enter(&vd->vdev_stat_lock); 3480b47119fdSGeorge Wilson if (type == ZIO_TYPE_READ && !vdev_is_dead(vd)) { 3481e14bb325SJeff Bonwick if (zio->io_error == ECKSUM) 3482e14bb325SJeff Bonwick vs->vs_checksum_errors++; 3483e14bb325SJeff Bonwick else 3484e14bb325SJeff Bonwick vs->vs_read_errors++; 3485fa9e4066Sahrens } 3486b47119fdSGeorge Wilson if (type == ZIO_TYPE_WRITE && !vdev_is_dead(vd)) 3487e14bb325SJeff Bonwick vs->vs_write_errors++; 3488e14bb325SJeff Bonwick mutex_exit(&vd->vdev_stat_lock); 3489fa9e4066Sahrens 34905cabbc6bSPrashanth Sreenivasa if (spa->spa_load_state == SPA_LOAD_NONE && 34915cabbc6bSPrashanth Sreenivasa type == ZIO_TYPE_WRITE && txg != 0 && 34928ad4d6ddSJeff Bonwick (!(flags & ZIO_FLAG_IO_REPAIR) || 349344ecc532SGeorge Wilson (flags & ZIO_FLAG_SCAN_THREAD) || 3494b24ab676SJeff Bonwick spa->spa_claiming)) { 34958ad4d6ddSJeff Bonwick /* 3496b24ab676SJeff Bonwick * This is either a normal write (not a repair), or it's 3497b24ab676SJeff Bonwick * a repair induced by the scrub thread, or it's a repair 3498b24ab676SJeff Bonwick * made by zil_claim() during spa_load() in the first txg. 3499b24ab676SJeff Bonwick * In the normal case, we commit the DTL change in the same 3500b24ab676SJeff Bonwick * txg as the block was born. In the scrub-induced repair 3501b24ab676SJeff Bonwick * case, we know that scrubs run in first-pass syncing context, 3502b24ab676SJeff Bonwick * so we commit the DTL change in spa_syncing_txg(spa). 3503b24ab676SJeff Bonwick * In the zil_claim() case, we commit in spa_first_txg(spa). 35048ad4d6ddSJeff Bonwick * 35058ad4d6ddSJeff Bonwick * We currently do not make DTL entries for failed spontaneous 35068ad4d6ddSJeff Bonwick * self-healing writes triggered by normal (non-scrubbing) 35078ad4d6ddSJeff Bonwick * reads, because we have no transactional context in which to 35088ad4d6ddSJeff Bonwick * do so -- and it's not clear that it'd be desirable anyway. 35098ad4d6ddSJeff Bonwick */ 35108ad4d6ddSJeff Bonwick if (vd->vdev_ops->vdev_op_leaf) { 35118ad4d6ddSJeff Bonwick uint64_t commit_txg = txg; 351244ecc532SGeorge Wilson if (flags & ZIO_FLAG_SCAN_THREAD) { 35138ad4d6ddSJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 35148ad4d6ddSJeff Bonwick ASSERT(spa_sync_pass(spa) == 1); 35158ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_SCRUB, txg, 1); 3516b24ab676SJeff Bonwick commit_txg = spa_syncing_txg(spa); 3517b24ab676SJeff Bonwick } else if (spa->spa_claiming) { 3518b24ab676SJeff Bonwick ASSERT(flags & ZIO_FLAG_IO_REPAIR); 3519b24ab676SJeff Bonwick commit_txg = spa_first_txg(spa); 35208ad4d6ddSJeff Bonwick } 3521b24ab676SJeff Bonwick ASSERT(commit_txg >= spa_syncing_txg(spa)); 35228ad4d6ddSJeff Bonwick if (vdev_dtl_contains(vd, DTL_MISSING, txg, 1)) 3523fa9e4066Sahrens return; 35248ad4d6ddSJeff Bonwick for (pvd = vd; pvd != rvd; pvd = pvd->vdev_parent) 35258ad4d6ddSJeff Bonwick vdev_dtl_dirty(pvd, DTL_PARTIAL, txg, 1); 35268ad4d6ddSJeff Bonwick vdev_dirty(vd->vdev_top, VDD_DTL, vd, commit_txg); 3527fa9e4066Sahrens } 35288ad4d6ddSJeff Bonwick if (vd != rvd) 35298ad4d6ddSJeff Bonwick vdev_dtl_dirty(vd, DTL_MISSING, txg, 1); 3530fa9e4066Sahrens } 3531fa9e4066Sahrens } 3532fa9e4066Sahrens 3533fa9e4066Sahrens /* 3534b24ab676SJeff Bonwick * Update the in-core space usage stats for this vdev, its metaslab class, 3535b24ab676SJeff Bonwick * and the root vdev. 3536fa9e4066Sahrens */ 3537fa9e4066Sahrens void 3538b24ab676SJeff Bonwick vdev_space_update(vdev_t *vd, int64_t alloc_delta, int64_t defer_delta, 3539b24ab676SJeff Bonwick int64_t space_delta) 3540fa9e4066Sahrens { 354199653d4eSeschrock int64_t dspace_delta = space_delta; 35428654d025Sperrin spa_t *spa = vd->vdev_spa; 35438654d025Sperrin vdev_t *rvd = spa->spa_root_vdev; 3544b24ab676SJeff Bonwick metaslab_group_t *mg = vd->vdev_mg; 3545b24ab676SJeff Bonwick metaslab_class_t *mc = mg ? mg->mg_class : NULL; 3546fa9e4066Sahrens 35478654d025Sperrin ASSERT(vd == vd->vdev_top); 354899653d4eSeschrock 35498654d025Sperrin /* 35508654d025Sperrin * Apply the inverse of the psize-to-asize (ie. RAID-Z) space-expansion 35518654d025Sperrin * factor. We must calculate this here and not at the root vdev 35528654d025Sperrin * because the root vdev's psize-to-asize is simply the max of its 35538654d025Sperrin * childrens', thus not accurate enough for us. 35548654d025Sperrin */ 35558654d025Sperrin ASSERT((dspace_delta & (SPA_MINBLOCKSIZE-1)) == 0); 3556e6ca193dSGeorge Wilson ASSERT(vd->vdev_deflate_ratio != 0 || vd->vdev_isl2cache); 35578654d025Sperrin dspace_delta = (dspace_delta >> SPA_MINBLOCKSHIFT) * 35588654d025Sperrin vd->vdev_deflate_ratio; 35598654d025Sperrin 35608654d025Sperrin mutex_enter(&vd->vdev_stat_lock); 35618654d025Sperrin vd->vdev_stat.vs_alloc += alloc_delta; 3562b24ab676SJeff Bonwick vd->vdev_stat.vs_space += space_delta; 35638654d025Sperrin vd->vdev_stat.vs_dspace += dspace_delta; 35648654d025Sperrin mutex_exit(&vd->vdev_stat_lock); 35658654d025Sperrin 3566b24ab676SJeff Bonwick if (mc == spa_normal_class(spa)) { 3567fa94a07fSbrendan mutex_enter(&rvd->vdev_stat_lock); 3568fa94a07fSbrendan rvd->vdev_stat.vs_alloc += alloc_delta; 3569b24ab676SJeff Bonwick rvd->vdev_stat.vs_space += space_delta; 3570fa94a07fSbrendan rvd->vdev_stat.vs_dspace += dspace_delta; 3571fa94a07fSbrendan mutex_exit(&rvd->vdev_stat_lock); 3572fa94a07fSbrendan } 3573b24ab676SJeff Bonwick 3574b24ab676SJeff Bonwick if (mc != NULL) { 3575b24ab676SJeff Bonwick ASSERT(rvd == vd->vdev_parent); 3576b24ab676SJeff Bonwick ASSERT(vd->vdev_ms_count != 0); 3577b24ab676SJeff Bonwick 3578b24ab676SJeff Bonwick metaslab_class_space_update(mc, 3579b24ab676SJeff Bonwick alloc_delta, defer_delta, space_delta, dspace_delta); 3580b24ab676SJeff Bonwick } 3581fa9e4066Sahrens } 3582fa9e4066Sahrens 3583fa9e4066Sahrens /* 3584fa9e4066Sahrens * Mark a top-level vdev's config as dirty, placing it on the dirty list 3585fa9e4066Sahrens * so that it will be written out next time the vdev configuration is synced. 3586fa9e4066Sahrens * If the root vdev is specified (vdev_top == NULL), dirty all top-level vdevs. 3587fa9e4066Sahrens */ 3588fa9e4066Sahrens void 3589fa9e4066Sahrens vdev_config_dirty(vdev_t *vd) 3590fa9e4066Sahrens { 3591fa9e4066Sahrens spa_t *spa = vd->vdev_spa; 3592fa9e4066Sahrens vdev_t *rvd = spa->spa_root_vdev; 3593fa9e4066Sahrens int c; 3594fa9e4066Sahrens 3595f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 3596f9af39baSGeorge Wilson 3597c5904d13Seschrock /* 35986809eb4eSEric Schrock * If this is an aux vdev (as with l2cache and spare devices), then we 35996809eb4eSEric Schrock * update the vdev config manually and set the sync flag. 3600c5904d13Seschrock */ 3601c5904d13Seschrock if (vd->vdev_aux != NULL) { 3602c5904d13Seschrock spa_aux_vdev_t *sav = vd->vdev_aux; 3603c5904d13Seschrock nvlist_t **aux; 3604c5904d13Seschrock uint_t naux; 3605c5904d13Seschrock 3606c5904d13Seschrock for (c = 0; c < sav->sav_count; c++) { 3607c5904d13Seschrock if (sav->sav_vdevs[c] == vd) 3608c5904d13Seschrock break; 3609c5904d13Seschrock } 3610c5904d13Seschrock 3611e14bb325SJeff Bonwick if (c == sav->sav_count) { 3612e14bb325SJeff Bonwick /* 3613e14bb325SJeff Bonwick * We're being removed. There's nothing more to do. 3614e14bb325SJeff Bonwick */ 3615e14bb325SJeff Bonwick ASSERT(sav->sav_sync == B_TRUE); 3616e14bb325SJeff Bonwick return; 3617e14bb325SJeff Bonwick } 3618e14bb325SJeff Bonwick 3619c5904d13Seschrock sav->sav_sync = B_TRUE; 3620c5904d13Seschrock 36216809eb4eSEric Schrock if (nvlist_lookup_nvlist_array(sav->sav_config, 36226809eb4eSEric Schrock ZPOOL_CONFIG_L2CACHE, &aux, &naux) != 0) { 36236809eb4eSEric Schrock VERIFY(nvlist_lookup_nvlist_array(sav->sav_config, 36246809eb4eSEric Schrock ZPOOL_CONFIG_SPARES, &aux, &naux) == 0); 36256809eb4eSEric Schrock } 3626c5904d13Seschrock 3627c5904d13Seschrock ASSERT(c < naux); 3628c5904d13Seschrock 3629c5904d13Seschrock /* 3630c5904d13Seschrock * Setting the nvlist in the middle if the array is a little 3631c5904d13Seschrock * sketchy, but it will work. 3632c5904d13Seschrock */ 3633c5904d13Seschrock nvlist_free(aux[c]); 36343f9d6ad7SLin Ling aux[c] = vdev_config_generate(spa, vd, B_TRUE, 0); 3635c5904d13Seschrock 3636c5904d13Seschrock return; 3637c5904d13Seschrock } 3638c5904d13Seschrock 36395dabedeeSbonwick /* 3640e14bb325SJeff Bonwick * The dirty list is protected by the SCL_CONFIG lock. The caller 3641e14bb325SJeff Bonwick * must either hold SCL_CONFIG as writer, or must be the sync thread 3642e14bb325SJeff Bonwick * (which holds SCL_CONFIG as reader). There's only one sync thread, 36435dabedeeSbonwick * so this is sufficient to ensure mutual exclusion. 36445dabedeeSbonwick */ 3645e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3646e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3647e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 36485dabedeeSbonwick 3649fa9e4066Sahrens if (vd == rvd) { 3650fa9e4066Sahrens for (c = 0; c < rvd->vdev_children; c++) 3651fa9e4066Sahrens vdev_config_dirty(rvd->vdev_child[c]); 3652fa9e4066Sahrens } else { 3653fa9e4066Sahrens ASSERT(vd == vd->vdev_top); 3654fa9e4066Sahrens 365588ecc943SGeorge Wilson if (!list_link_active(&vd->vdev_config_dirty_node) && 36565cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) { 3657e14bb325SJeff Bonwick list_insert_head(&spa->spa_config_dirty_list, vd); 36585cabbc6bSPrashanth Sreenivasa } 3659fa9e4066Sahrens } 3660fa9e4066Sahrens } 3661fa9e4066Sahrens 3662fa9e4066Sahrens void 3663fa9e4066Sahrens vdev_config_clean(vdev_t *vd) 3664fa9e4066Sahrens { 36655dabedeeSbonwick spa_t *spa = vd->vdev_spa; 36665dabedeeSbonwick 3667e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_CONFIG, RW_WRITER) || 3668e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3669e14bb325SJeff Bonwick spa_config_held(spa, SCL_CONFIG, RW_READER))); 36705dabedeeSbonwick 3671e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_config_dirty_node)); 3672e14bb325SJeff Bonwick list_remove(&spa->spa_config_dirty_list, vd); 3673e14bb325SJeff Bonwick } 3674e14bb325SJeff Bonwick 3675e14bb325SJeff Bonwick /* 3676e14bb325SJeff Bonwick * Mark a top-level vdev's state as dirty, so that the next pass of 3677e14bb325SJeff Bonwick * spa_sync() can convert this into vdev_config_dirty(). We distinguish 3678e14bb325SJeff Bonwick * the state changes from larger config changes because they require 3679e14bb325SJeff Bonwick * much less locking, and are often needed for administrative actions. 3680e14bb325SJeff Bonwick */ 3681e14bb325SJeff Bonwick void 3682e14bb325SJeff Bonwick vdev_state_dirty(vdev_t *vd) 3683e14bb325SJeff Bonwick { 3684e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3685e14bb325SJeff Bonwick 3686f9af39baSGeorge Wilson ASSERT(spa_writeable(spa)); 3687e14bb325SJeff Bonwick ASSERT(vd == vd->vdev_top); 3688e14bb325SJeff Bonwick 3689e14bb325SJeff Bonwick /* 3690e14bb325SJeff Bonwick * The state list is protected by the SCL_STATE lock. The caller 3691e14bb325SJeff Bonwick * must either hold SCL_STATE as writer, or must be the sync thread 3692e14bb325SJeff Bonwick * (which holds SCL_STATE as reader). There's only one sync thread, 3693e14bb325SJeff Bonwick * so this is sufficient to ensure mutual exclusion. 3694e14bb325SJeff Bonwick */ 3695e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3696e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3697e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3698e14bb325SJeff Bonwick 36995cabbc6bSPrashanth Sreenivasa if (!list_link_active(&vd->vdev_state_dirty_node) && 37005cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) 3701e14bb325SJeff Bonwick list_insert_head(&spa->spa_state_dirty_list, vd); 3702e14bb325SJeff Bonwick } 3703e14bb325SJeff Bonwick 3704e14bb325SJeff Bonwick void 3705e14bb325SJeff Bonwick vdev_state_clean(vdev_t *vd) 3706e14bb325SJeff Bonwick { 3707e14bb325SJeff Bonwick spa_t *spa = vd->vdev_spa; 3708e14bb325SJeff Bonwick 3709e14bb325SJeff Bonwick ASSERT(spa_config_held(spa, SCL_STATE, RW_WRITER) || 3710e14bb325SJeff Bonwick (dsl_pool_sync_context(spa_get_dsl(spa)) && 3711e14bb325SJeff Bonwick spa_config_held(spa, SCL_STATE, RW_READER))); 3712e14bb325SJeff Bonwick 3713e14bb325SJeff Bonwick ASSERT(list_link_active(&vd->vdev_state_dirty_node)); 3714e14bb325SJeff Bonwick list_remove(&spa->spa_state_dirty_list, vd); 3715fa9e4066Sahrens } 3716fa9e4066Sahrens 371732b87932Sek /* 371832b87932Sek * Propagate vdev state up from children to parent. 371932b87932Sek */ 372044cd46caSbillm void 372144cd46caSbillm vdev_propagate_state(vdev_t *vd) 372244cd46caSbillm { 37238ad4d6ddSJeff Bonwick spa_t *spa = vd->vdev_spa; 37248ad4d6ddSJeff Bonwick vdev_t *rvd = spa->spa_root_vdev; 372544cd46caSbillm int degraded = 0, faulted = 0; 372644cd46caSbillm int corrupted = 0; 372744cd46caSbillm vdev_t *child; 372844cd46caSbillm 37293d7072f8Seschrock if (vd->vdev_children > 0) { 3730573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 37313d7072f8Seschrock child = vd->vdev_child[c]; 373251ece835Seschrock 373388ecc943SGeorge Wilson /* 37345cabbc6bSPrashanth Sreenivasa * Don't factor holes or indirect vdevs into the 37355cabbc6bSPrashanth Sreenivasa * decision. 373688ecc943SGeorge Wilson */ 37375cabbc6bSPrashanth Sreenivasa if (!vdev_is_concrete(child)) 373888ecc943SGeorge Wilson continue; 373988ecc943SGeorge Wilson 3740e14bb325SJeff Bonwick if (!vdev_readable(child) || 37418ad4d6ddSJeff Bonwick (!vdev_writeable(child) && spa_writeable(spa))) { 374251ece835Seschrock /* 374351ece835Seschrock * Root special: if there is a top-level log 374451ece835Seschrock * device, treat the root vdev as if it were 374551ece835Seschrock * degraded. 374651ece835Seschrock */ 374751ece835Seschrock if (child->vdev_islog && vd == rvd) 374851ece835Seschrock degraded++; 374951ece835Seschrock else 375051ece835Seschrock faulted++; 375151ece835Seschrock } else if (child->vdev_state <= VDEV_STATE_DEGRADED) { 37523d7072f8Seschrock degraded++; 375351ece835Seschrock } 375444cd46caSbillm 37553d7072f8Seschrock if (child->vdev_stat.vs_aux == VDEV_AUX_CORRUPT_DATA) 37563d7072f8Seschrock corrupted++; 37573d7072f8Seschrock } 375844cd46caSbillm 37593d7072f8Seschrock vd->vdev_ops->vdev_op_state_change(vd, faulted, degraded); 37603d7072f8Seschrock 37613d7072f8Seschrock /* 3762e14bb325SJeff Bonwick * Root special: if there is a top-level vdev that cannot be 37633d7072f8Seschrock * opened due to corrupted metadata, then propagate the root 37643d7072f8Seschrock * vdev's aux state as 'corrupt' rather than 'insufficient 37653d7072f8Seschrock * replicas'. 37663d7072f8Seschrock */ 37673d7072f8Seschrock if (corrupted && vd == rvd && 37683d7072f8Seschrock rvd->vdev_state == VDEV_STATE_CANT_OPEN) 37693d7072f8Seschrock vdev_set_state(rvd, B_FALSE, VDEV_STATE_CANT_OPEN, 37703d7072f8Seschrock VDEV_AUX_CORRUPT_DATA); 37713d7072f8Seschrock } 37723d7072f8Seschrock 377351ece835Seschrock if (vd->vdev_parent) 37743d7072f8Seschrock vdev_propagate_state(vd->vdev_parent); 377544cd46caSbillm } 377644cd46caSbillm 3777fa9e4066Sahrens /* 3778ea8dc4b6Seschrock * Set a vdev's state. If this is during an open, we don't update the parent 3779ea8dc4b6Seschrock * state, because we're in the process of opening children depth-first. 3780ea8dc4b6Seschrock * Otherwise, we propagate the change to the parent. 3781ea8dc4b6Seschrock * 3782ea8dc4b6Seschrock * If this routine places a device in a faulted state, an appropriate ereport is 3783ea8dc4b6Seschrock * generated. 3784fa9e4066Sahrens */ 3785fa9e4066Sahrens void 3786ea8dc4b6Seschrock vdev_set_state(vdev_t *vd, boolean_t isopen, vdev_state_t state, vdev_aux_t aux) 3787fa9e4066Sahrens { 3788560e6e96Seschrock uint64_t save_state; 3789c5904d13Seschrock spa_t *spa = vd->vdev_spa; 3790ea8dc4b6Seschrock 3791ea8dc4b6Seschrock if (state == vd->vdev_state) { 3792ea8dc4b6Seschrock vd->vdev_stat.vs_aux = aux; 3793fa9e4066Sahrens return; 3794ea8dc4b6Seschrock } 3795ea8dc4b6Seschrock 3796560e6e96Seschrock save_state = vd->vdev_state; 3797fa9e4066Sahrens 3798fa9e4066Sahrens vd->vdev_state = state; 3799fa9e4066Sahrens vd->vdev_stat.vs_aux = aux; 3800fa9e4066Sahrens 38013d7072f8Seschrock /* 38023d7072f8Seschrock * If we are setting the vdev state to anything but an open state, then 380398d1cbfeSGeorge Wilson * always close the underlying device unless the device has requested 380498d1cbfeSGeorge Wilson * a delayed close (i.e. we're about to remove or fault the device). 380598d1cbfeSGeorge Wilson * Otherwise, we keep accessible but invalid devices open forever. 380698d1cbfeSGeorge Wilson * We don't call vdev_close() itself, because that implies some extra 380798d1cbfeSGeorge Wilson * checks (offline, etc) that we don't want here. This is limited to 380898d1cbfeSGeorge Wilson * leaf devices, because otherwise closing the device will affect other 380998d1cbfeSGeorge Wilson * children. 381098d1cbfeSGeorge Wilson */ 381198d1cbfeSGeorge Wilson if (!vd->vdev_delayed_close && vdev_is_dead(vd) && 381298d1cbfeSGeorge Wilson vd->vdev_ops->vdev_op_leaf) 38133d7072f8Seschrock vd->vdev_ops->vdev_op_close(vd); 38143d7072f8Seschrock 3815069f55e2SEric Schrock /* 3816069f55e2SEric Schrock * If we have brought this vdev back into service, we need 3817069f55e2SEric Schrock * to notify fmd so that it can gracefully repair any outstanding 3818069f55e2SEric Schrock * cases due to a missing device. We do this in all cases, even those 3819069f55e2SEric Schrock * that probably don't correlate to a repaired fault. This is sure to 3820069f55e2SEric Schrock * catch all cases, and we let the zfs-retire agent sort it out. If 3821069f55e2SEric Schrock * this is a transient state it's OK, as the retire agent will 3822069f55e2SEric Schrock * double-check the state of the vdev before repairing it. 3823069f55e2SEric Schrock */ 3824069f55e2SEric Schrock if (state == VDEV_STATE_HEALTHY && vd->vdev_ops->vdev_op_leaf && 3825069f55e2SEric Schrock vd->vdev_prevstate != state) 3826069f55e2SEric Schrock zfs_post_state_change(spa, vd); 3827069f55e2SEric Schrock 38283d7072f8Seschrock if (vd->vdev_removed && 38293d7072f8Seschrock state == VDEV_STATE_CANT_OPEN && 38303d7072f8Seschrock (aux == VDEV_AUX_OPEN_FAILED || vd->vdev_checkremove)) { 38313d7072f8Seschrock /* 38323d7072f8Seschrock * If the previous state is set to VDEV_STATE_REMOVED, then this 38333d7072f8Seschrock * device was previously marked removed and someone attempted to 38343d7072f8Seschrock * reopen it. If this failed due to a nonexistent device, then 38353d7072f8Seschrock * keep the device in the REMOVED state. We also let this be if 38363d7072f8Seschrock * it is one of our special test online cases, which is only 38373d7072f8Seschrock * attempting to online the device and shouldn't generate an FMA 38383d7072f8Seschrock * fault. 38393d7072f8Seschrock */ 38403d7072f8Seschrock vd->vdev_state = VDEV_STATE_REMOVED; 38413d7072f8Seschrock vd->vdev_stat.vs_aux = VDEV_AUX_NONE; 38423d7072f8Seschrock } else if (state == VDEV_STATE_REMOVED) { 38433d7072f8Seschrock vd->vdev_removed = B_TRUE; 38443d7072f8Seschrock } else if (state == VDEV_STATE_CANT_OPEN) { 3845ea8dc4b6Seschrock /* 3846cb04b873SMark J Musante * If we fail to open a vdev during an import or recovery, we 3847cb04b873SMark J Musante * mark it as "not available", which signifies that it was 3848cb04b873SMark J Musante * never there to begin with. Failure to open such a device 3849cb04b873SMark J Musante * is not considered an error. 3850ea8dc4b6Seschrock */ 3851cb04b873SMark J Musante if ((spa_load_state(spa) == SPA_LOAD_IMPORT || 3852cb04b873SMark J Musante spa_load_state(spa) == SPA_LOAD_RECOVER) && 3853560e6e96Seschrock vd->vdev_ops->vdev_op_leaf) 3854560e6e96Seschrock vd->vdev_not_present = 1; 3855560e6e96Seschrock 3856560e6e96Seschrock /* 3857560e6e96Seschrock * Post the appropriate ereport. If the 'prevstate' field is 3858560e6e96Seschrock * set to something other than VDEV_STATE_UNKNOWN, it indicates 3859560e6e96Seschrock * that this is part of a vdev_reopen(). In this case, we don't 3860560e6e96Seschrock * want to post the ereport if the device was already in the 3861560e6e96Seschrock * CANT_OPEN state beforehand. 38623d7072f8Seschrock * 38633d7072f8Seschrock * If the 'checkremove' flag is set, then this is an attempt to 38643d7072f8Seschrock * online the device in response to an insertion event. If we 38653d7072f8Seschrock * hit this case, then we have detected an insertion event for a 38663d7072f8Seschrock * faulted or offline device that wasn't in the removed state. 38673d7072f8Seschrock * In this scenario, we don't post an ereport because we are 38683d7072f8Seschrock * about to replace the device, or attempt an online with 38693d7072f8Seschrock * vdev_forcefault, which will generate the fault for us. 3870560e6e96Seschrock */ 38713d7072f8Seschrock if ((vd->vdev_prevstate != state || vd->vdev_forcefault) && 38723d7072f8Seschrock !vd->vdev_not_present && !vd->vdev_checkremove && 3873c5904d13Seschrock vd != spa->spa_root_vdev) { 3874ea8dc4b6Seschrock const char *class; 3875ea8dc4b6Seschrock 3876ea8dc4b6Seschrock switch (aux) { 3877ea8dc4b6Seschrock case VDEV_AUX_OPEN_FAILED: 3878ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_OPEN_FAILED; 3879ea8dc4b6Seschrock break; 3880ea8dc4b6Seschrock case VDEV_AUX_CORRUPT_DATA: 3881ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_CORRUPT_DATA; 3882ea8dc4b6Seschrock break; 3883ea8dc4b6Seschrock case VDEV_AUX_NO_REPLICAS: 3884ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_NO_REPLICAS; 3885ea8dc4b6Seschrock break; 3886ea8dc4b6Seschrock case VDEV_AUX_BAD_GUID_SUM: 3887ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_GUID_SUM; 3888ea8dc4b6Seschrock break; 3889ea8dc4b6Seschrock case VDEV_AUX_TOO_SMALL: 3890ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_TOO_SMALL; 3891ea8dc4b6Seschrock break; 3892ea8dc4b6Seschrock case VDEV_AUX_BAD_LABEL: 3893ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_BAD_LABEL; 3894ea8dc4b6Seschrock break; 3895ea8dc4b6Seschrock default: 3896ea8dc4b6Seschrock class = FM_EREPORT_ZFS_DEVICE_UNKNOWN; 3897ea8dc4b6Seschrock } 3898ea8dc4b6Seschrock 3899c5904d13Seschrock zfs_ereport_post(class, spa, vd, NULL, save_state, 0); 3900ea8dc4b6Seschrock } 3901ea8dc4b6Seschrock 39023d7072f8Seschrock /* Erase any notion of persistent removed state */ 39033d7072f8Seschrock vd->vdev_removed = B_FALSE; 39043d7072f8Seschrock } else { 39053d7072f8Seschrock vd->vdev_removed = B_FALSE; 39063d7072f8Seschrock } 3907ea8dc4b6Seschrock 39088b33d774STim Haley if (!isopen && vd->vdev_parent) 39098b33d774STim Haley vdev_propagate_state(vd->vdev_parent); 3910fa9e4066Sahrens } 391115e6edf1Sgw 39126f793812SPavel Zakharov boolean_t 39136f793812SPavel Zakharov vdev_children_are_offline(vdev_t *vd) 39146f793812SPavel Zakharov { 39156f793812SPavel Zakharov ASSERT(!vd->vdev_ops->vdev_op_leaf); 39166f793812SPavel Zakharov 39176f793812SPavel Zakharov for (uint64_t i = 0; i < vd->vdev_children; i++) { 39186f793812SPavel Zakharov if (vd->vdev_child[i]->vdev_state != VDEV_STATE_OFFLINE) 39196f793812SPavel Zakharov return (B_FALSE); 39206f793812SPavel Zakharov } 39216f793812SPavel Zakharov 39226f793812SPavel Zakharov return (B_TRUE); 39236f793812SPavel Zakharov } 39246f793812SPavel Zakharov 392515e6edf1Sgw /* 392615e6edf1Sgw * Check the vdev configuration to ensure that it's capable of supporting 3927c8811bd3SToomas Soome * a root pool. We do not support partial configuration. 3928c8811bd3SToomas Soome * In addition, only a single top-level vdev is allowed. 392915e6edf1Sgw */ 393015e6edf1Sgw boolean_t 393115e6edf1Sgw vdev_is_bootable(vdev_t *vd) 393215e6edf1Sgw { 393315e6edf1Sgw if (!vd->vdev_ops->vdev_op_leaf) { 393415e6edf1Sgw char *vdev_type = vd->vdev_ops->vdev_op_type; 393515e6edf1Sgw 393615e6edf1Sgw if (strcmp(vdev_type, VDEV_TYPE_ROOT) == 0 && 393715e6edf1Sgw vd->vdev_children > 1) { 393815e6edf1Sgw return (B_FALSE); 39395cabbc6bSPrashanth Sreenivasa } else if (strcmp(vdev_type, VDEV_TYPE_MISSING) == 0 || 39405cabbc6bSPrashanth Sreenivasa strcmp(vdev_type, VDEV_TYPE_INDIRECT) == 0) { 394115e6edf1Sgw return (B_FALSE); 394215e6edf1Sgw } 394315e6edf1Sgw } 394415e6edf1Sgw 3945573ca77eSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) { 394615e6edf1Sgw if (!vdev_is_bootable(vd->vdev_child[c])) 394715e6edf1Sgw return (B_FALSE); 394815e6edf1Sgw } 394915e6edf1Sgw return (B_TRUE); 395015e6edf1Sgw } 3951e6ca193dSGeorge Wilson 39525cabbc6bSPrashanth Sreenivasa boolean_t 39535cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vdev_t *vd) 39545cabbc6bSPrashanth Sreenivasa { 39555cabbc6bSPrashanth Sreenivasa vdev_ops_t *ops = vd->vdev_ops; 39565cabbc6bSPrashanth Sreenivasa if (ops == &vdev_indirect_ops || ops == &vdev_hole_ops || 39575cabbc6bSPrashanth Sreenivasa ops == &vdev_missing_ops || ops == &vdev_root_ops) { 39585cabbc6bSPrashanth Sreenivasa return (B_FALSE); 39595cabbc6bSPrashanth Sreenivasa } else { 39605cabbc6bSPrashanth Sreenivasa return (B_TRUE); 39615cabbc6bSPrashanth Sreenivasa } 39625cabbc6bSPrashanth Sreenivasa } 39635cabbc6bSPrashanth Sreenivasa 39644b964adaSGeorge Wilson /* 39654b964adaSGeorge Wilson * Determine if a log device has valid content. If the vdev was 39664b964adaSGeorge Wilson * removed or faulted in the MOS config then we know that 39674b964adaSGeorge Wilson * the content on the log device has already been written to the pool. 39684b964adaSGeorge Wilson */ 39694b964adaSGeorge Wilson boolean_t 39704b964adaSGeorge Wilson vdev_log_state_valid(vdev_t *vd) 39714b964adaSGeorge Wilson { 39724b964adaSGeorge Wilson if (vd->vdev_ops->vdev_op_leaf && !vd->vdev_faulted && 39734b964adaSGeorge Wilson !vd->vdev_removed) 39744b964adaSGeorge Wilson return (B_TRUE); 39754b964adaSGeorge Wilson 39764b964adaSGeorge Wilson for (int c = 0; c < vd->vdev_children; c++) 39774b964adaSGeorge Wilson if (vdev_log_state_valid(vd->vdev_child[c])) 39784b964adaSGeorge Wilson return (B_TRUE); 39794b964adaSGeorge Wilson 39804b964adaSGeorge Wilson return (B_FALSE); 39814b964adaSGeorge Wilson } 39824b964adaSGeorge Wilson 3983573ca77eSGeorge Wilson /* 3984573ca77eSGeorge Wilson * Expand a vdev if possible. 3985573ca77eSGeorge Wilson */ 3986573ca77eSGeorge Wilson void 3987573ca77eSGeorge Wilson vdev_expand(vdev_t *vd, uint64_t txg) 3988573ca77eSGeorge Wilson { 3989573ca77eSGeorge Wilson ASSERT(vd->vdev_top == vd); 3990573ca77eSGeorge Wilson ASSERT(spa_config_held(vd->vdev_spa, SCL_ALL, RW_WRITER) == SCL_ALL); 3991573ca77eSGeorge Wilson 39925cabbc6bSPrashanth Sreenivasa vdev_set_deflate_ratio(vd); 39935cabbc6bSPrashanth Sreenivasa 39945cabbc6bSPrashanth Sreenivasa if ((vd->vdev_asize >> vd->vdev_ms_shift) > vd->vdev_ms_count && 39955cabbc6bSPrashanth Sreenivasa vdev_is_concrete(vd)) { 3996573ca77eSGeorge Wilson VERIFY(vdev_metaslab_init(vd, txg) == 0); 3997573ca77eSGeorge Wilson vdev_config_dirty(vd); 3998573ca77eSGeorge Wilson } 3999573ca77eSGeorge Wilson } 40001195e687SMark J Musante 40011195e687SMark J Musante /* 40021195e687SMark J Musante * Split a vdev. 40031195e687SMark J Musante */ 40041195e687SMark J Musante void 40051195e687SMark J Musante vdev_split(vdev_t *vd) 40061195e687SMark J Musante { 40071195e687SMark J Musante vdev_t *cvd, *pvd = vd->vdev_parent; 40081195e687SMark J Musante 40091195e687SMark J Musante vdev_remove_child(pvd, vd); 40101195e687SMark J Musante vdev_compact_children(pvd); 40111195e687SMark J Musante 40121195e687SMark J Musante cvd = pvd->vdev_child[0]; 40131195e687SMark J Musante if (pvd->vdev_children == 1) { 40141195e687SMark J Musante vdev_remove_parent(cvd); 40151195e687SMark J Musante cvd->vdev_splitting = B_TRUE; 40161195e687SMark J Musante } 40171195e687SMark J Musante vdev_propagate_state(cvd); 40181195e687SMark J Musante } 4019283b8460SGeorge.Wilson 4020283b8460SGeorge.Wilson void 4021283b8460SGeorge.Wilson vdev_deadman(vdev_t *vd) 4022283b8460SGeorge.Wilson { 4023283b8460SGeorge.Wilson for (int c = 0; c < vd->vdev_children; c++) { 4024283b8460SGeorge.Wilson vdev_t *cvd = vd->vdev_child[c]; 4025283b8460SGeorge.Wilson 4026283b8460SGeorge.Wilson vdev_deadman(cvd); 4027283b8460SGeorge.Wilson } 4028283b8460SGeorge.Wilson 4029283b8460SGeorge.Wilson if (vd->vdev_ops->vdev_op_leaf) { 4030283b8460SGeorge.Wilson vdev_queue_t *vq = &vd->vdev_queue; 4031283b8460SGeorge.Wilson 4032283b8460SGeorge.Wilson mutex_enter(&vq->vq_lock); 403369962b56SMatthew Ahrens if (avl_numnodes(&vq->vq_active_tree) > 0) { 4034283b8460SGeorge.Wilson spa_t *spa = vd->vdev_spa; 4035283b8460SGeorge.Wilson zio_t *fio; 4036283b8460SGeorge.Wilson uint64_t delta; 4037283b8460SGeorge.Wilson 4038283b8460SGeorge.Wilson /* 4039283b8460SGeorge.Wilson * Look at the head of all the pending queues, 4040283b8460SGeorge.Wilson * if any I/O has been outstanding for longer than 4041283b8460SGeorge.Wilson * the spa_deadman_synctime we panic the system. 4042283b8460SGeorge.Wilson */ 404369962b56SMatthew Ahrens fio = avl_first(&vq->vq_active_tree); 4044c55e05cbSMatthew Ahrens delta = gethrtime() - fio->io_timestamp; 4045c55e05cbSMatthew Ahrens if (delta > spa_deadman_synctime(spa)) { 40463ee8c80cSPavel Zakharov vdev_dbgmsg(vd, "SLOW IO: zio timestamp " 40473ee8c80cSPavel Zakharov "%lluns, delta %lluns, last io %lluns", 40483ee8c80cSPavel Zakharov fio->io_timestamp, (u_longlong_t)delta, 4049283b8460SGeorge.Wilson vq->vq_io_complete_ts); 4050283b8460SGeorge.Wilson fm_panic("I/O to pool '%s' appears to be " 4051283b8460SGeorge.Wilson "hung.", spa_name(spa)); 4052283b8460SGeorge.Wilson } 4053283b8460SGeorge.Wilson } 4054283b8460SGeorge.Wilson mutex_exit(&vq->vq_lock); 4055283b8460SGeorge.Wilson } 4056283b8460SGeorge.Wilson } 4057